Feature/tle propagation

[SuperdoerTrav]

Make TLE propagation faithful, add drag fitting, and fix a frame approximation

What this does

When you load a TLE (two-line element set) into SSAPy today, the drag
information in it is thrown away, so propagating a TLE drifts away from the
standard SGP4 result — about 7 km/day for the ISS. This PR fixes that and adds
two related capabilities, plus a documentation refresh. Each of the five commits
stands on its own.

Three things change:

1. TLEs now propagate exactly like standard SGP4. Load a TLE, get the same
   answer the reference SGP4 library gives — including the drag term.
2. A new physics-based option for TLEs, with orbit fitting. Instead of SGP4,
   you can propagate a TLE through SSAPy's full numerical force model and fit the
   satellite's drag (and, if you want, its state) to real tracking data.
3. A more accurate coordinate-frame conversion in the SGP4 propagator.

---

1. Faithful SGP4 from a TLE

A TLE carries a drag term (B*) and mean-motion rates. SSAPy was discarding
them and propagating from drag-free elements, so results diverged from SGP4 —
worst for low, high-drag objects.

Fix: keep the original SGP4 record when a TLE is loaded and propagate with
it directly. Orbit.fromTLETuple now stores the parsed SGP4 record; the SGP4
propagator uses it when present (the old reconstruction path stays as a
fallback). The record is preserved when an orbit is indexed or iterated, which
is where it was previously being dropped.

Result — position error vs the reference SGP4 library:

object	before	after
ISS (low Earth orbit, ~420 km)	7153 m at 24 h	~1e-9 m
geostationary	~20 m at 24 h	~1e-9 m

This makes a TLE-loaded orbit a faithful SGP4 baseline. Note it is running
SGP4, not independently reproducing it.

---

2. Physics-based propagation and drag fitting for TLEs (ssapy.tle_drag)

The alternative to SGP4: take a TLE as a starting point and propagate it through
SSAPy's numerical force model (gravity + Sun/Moon + atmospheric drag). This is a
genuinely different, higher-fidelity model — it does not reproduce SGP4 and
should be checked against real tracking data, not against SGP4.

A TLE doesn't give you a physical drag coefficient (its B* is tied to SGP4's
internal atmosphere), so this module estimates one:

• fit_drag — solves for the satellite's drag coefficient (Cd·A/m) so the
  numerical propagation matches a reference orbit track.
• fit_orbit_drag — a full orbit-determination fit that solves for the
  satellite's position, velocity, and drag together. This matters because if
  the starting position is even slightly off and you only fit drag, the position
  error gets absorbed into the drag number and corrupts it. Fitting everything at
  once separates them.

Demonstration on a known test case (starting position deliberately off by ~210 m
and 0.18 m/s; true drag coefficient 5e-3):

approach	recovered drag coefficient	track fit
fit drag only (position held fixed)	0.327 — off by +6400%	stuck at ~1.7 km error
fit position + drag together	5.0e-3 — essentially exact	~1e-7 m error

Helpers included: convert B* to a starting drag estimate, build the numerical
propagator, and (optionally) return the fit's uncertainty covariance for
downstream filtering. fit_drag was also sped up to integrate an orbit once
across all requested times instead of re-integrating for each one.

---

3. Time-accurate TEME→GCRF conversion

SGP4 produces coordinates in the TEME frame; SSAPy converts them to the standard
GCRF/GCRS frame. The propagator was computing that rotation once, at the start
time, and reusing it for the whole arc. Because the rotation drifts slowly over
time, this leaves a small error that grows — and grows with orbit size.

Fix: compute the rotation at each output time. Checked against Astropy's
independent implementation:

object	old (single time) at 12 h	new (per time step)
ISS	2.2 m	~0.01 m
geostationary	13.5 m (26.5 m at 24 h)	~0.05 m

Also fixes the underlying conversion so it accepts an array of times (it silently
failed on arrays before); single-time results are bit-for-bit identical.

Behavior change for existing users: SGP4Propagator() now does the
per-time-step conversion by default. Output shifts by up to a few tens of
meters at geostationary altitude compared to before. This is the more correct
behavior. To get the old single-time behavior, pass the epoch:
SGP4Propagator(t=<epoch>).

---

Runs offline

No new runtime network dependencies. Earth-orientation data comes from Astropy's
bundled table (no downloads, safely clamps out-of-range dates); the numerical
model uses SSAPy's bundled ephemeris/gravity data; the fits are plain
NumPy/SciPy. The drag model stays Harris-Priester, which needs no external
space-weather data.

Documentation

• Cite the published JOSS paper (Meyers et al. 2025, JOSS 10(111), 8147,
  doi:10.21105/joss.08147) as the primary citation, with BibTeX, and point the
  JOSS badge at the DOI.
• Add a dedicated SSAPy-Toolkit section describing the companion project.
• Fix a broken link in the README (it pointed at a bib file that doesn't exist).

Testing

• New: tests/test_tle_drag.py (8 tests — exact-SGP4 match, drag fit, joint
  orbit+drag fit) and tests/test_frame_perstep.py (3 tests — array support and
  accuracy vs Astropy).
• Existing test_frame.py, test_orbit.py, test_io.py: 27 pass, no
  regressions.
• examples/tle_drag_paths.py demonstrates all three features end to end.

Possible follow-ups (not in this PR)

• Faster/more-robust Jacobian for the orbit-determination fit (currently
  finite-difference).
• Optional higher-fidelity atmosphere model (NRLMSISE-00 / JB2008) as a drop-in
  drag option — deferred to keep SSAPy fully offline for now.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 22.98137% with 124 lines in your changes missing coverage. Please review.
✅ Project coverage is 58.81%. Comparing base (ddc4cf9) to head (a7960ee).

Files with missing lines	Patch %	Lines
ssapy/tle_drag.py	8.27%	122 Missing ⚠️
ssapy/orbit.py	81.82%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #115      +/-   ##
==========================================
- Coverage   59.89%   58.81%   -1.07%     
==========================================
  Files          17       18       +1     
  Lines        5574     5730     +156     
==========================================
+ Hits         3338     3370      +32     
- Misses       2236     2360     +124     

Flag	Coverage Δ
unittests	58.81% <22.98%> (-1.07%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files with missing lines	Coverage Δ
ssapy/compute.py	57.79% <100.00%> (+0.54%)	⬆️
ssapy/propagator.py	86.11% <100.00%> (+0.13%)	⬆️
ssapy/utils.py	53.17% <100.00%> (+0.12%)	⬆️
ssapy/orbit.py	84.89% <81.82%> (-0.04%)	⬇️
ssapy/tle_drag.py	8.27% <8.27%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.