Guard the Dask scatter in run_TPI so serial (client=None) runs work

[vahid-ahmadi]

Why

run_TPI(p, client=None) crashes. client.scatter is called unconditionally, so passing no Dask client raises AttributeError before the TPI loop is entered — which makes the serial fallback at TPI.py:1039 unreachable in exactly the case it exists for. SS.inner_loop already guards the equivalent block with if client:; TPI doesn't.

What changes after merging

Serial TPI runs work. Anyone calling run_TPI without a Dask cluster — scripts, notebooks, CI, small debugging runs — goes from a hard crash to a working solve. No behaviour changes when a client is supplied.

Change

Wraps the schema-backup / scatter / restore block in if client:. scattered_p_future is only read inside the if client: submit branch, so this is a guard plus an indent — no logic moves.

Evidence

New tests/test_TPI.py::test_run_TPI_serial_no_client seeds SS from the cached test_io_data pickles and monkeypatches inner_loop to raise a sentinel, asserting execution reaches the first serial household solve. No solve is run; ~1s, no Dask, so it is unmarked and guards this path in normal CI.

with the fix:      1 passed in 1.15s
fix stashed:       FAILED - AttributeError: 'NoneType' object has no attribute 'scatter'

Fixes the first half of #1211. The second half (SS re-scattering per residual evaluation) is #1214.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 8.33333% with 11 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.71%. Comparing base (a30defc) to head (9a2f823).

Files with missing lines	Patch %	Lines
ogcore/TPI.py	8.33%	11 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1212      +/-   ##
==========================================
+ Coverage   74.07%   75.71%   +1.64%     
==========================================
  Files          22       22              
  Lines        5920     5921       +1     
==========================================
+ Hits         4385     4483      +98     
+ Misses       1535     1438      -97     

Flag	Coverage Δ
unittests	75.71% <8.33%> (+1.64%)	⬆️

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

Files with missing lines	Coverage Δ
ogcore/TPI.py	52.42% <8.33%> (+17.17%)	⬆️

... and 1 file with indirect coverage changes

🚀 New features to boost your workflow:

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

[vahid-ahmadi]

On the Codecov flag (8.33% patch coverage, 11 lines missing) — worth explaining, since it reads worse than it is.

The diff is a guard plus a re-indent, so Codecov counts the 11 re-indented lines of the schema-backup/scatter/restore block as new. Those lines only execute with a Dask client, and the non-local CI suite runs without one — so they show as uncovered even though they are byte-identical to what was already there and already ran in the local-marked tests.

The same report shows ogcore/TPI.py coverage going up 17.17% (35.25% → 52.42%), which is the real effect: before this change run_TPI(p, client=None) raised AttributeError before the loop, so the entire serial path was unreachable and uncovered. It now executes.

I have separately verified the serial path is not just reachable but numerically correct: with ENFORCE_SOLUTION_CHECKS = False, serial output compared against a real LocalCluster run across all 72 keys of the returned dict gives 0 mismatches at 1e-8. The regression test in this PR asserts reachability only, which is the narrow thing that was broken — I did not want to add a multi-minute solve to CI for it.