Remove unreachable checks and improve test coverage

[jgabry]

Submission Checklist

• Run unit tests
• Declare copyright holder and agree to license (see below)

Summary

Part of #1258. Before the v1.0 release candidate we went through every
line Codecov reports as uncovered (Linux and Windows sessions merged) and
decided each one: a test, a removal, or an accepted gap.

Removed, because nothing can reach them:

• validate_exe_file(): every method that runs the executable calls the
  model's assert_current() first, which already reports a missing
  executable.
• The format_version check in validate_build_record():
  read_build_record() checks the version before calling the validator
  and new_build_record() writes it.
• The rvars cases in the draws-format helpers: assert_valid_draws_format()
  rejects that format first. $unconstrain_draws() now uses that assertion
  instead of its own check.
• Two branches of .remove_leftmost_dim() for shapes draws_of() never
  produces, and the unused show_stderr_messages() getter.

The guard against reading a fit whose chains all failed was repeated in
six public methods with three different messages. It now lives only in
each class's read_csv_(), so $draws(), $summary(), $print(),
$metadata(), $sampler_diagnostics() and $inv_metric() give the same
error, and $draws() no longer errors when the draws are already cached.
No NEWS entry: only the wording of that error changes.

New tests go through the public API: iter_warmup = 0 with
save_warmup = TRUE, a fit as init that failed or has different
parameters, $diagnostic_summary(quiet = TRUE), $loo(r_eff = TRUE),
$print(line_numbers = TRUE), format_stan_file() with canonicalize = TRUE and max_line_length, the release-candidate branch of
cmdstan_default_path(), a makefile without a version line, the knitr
engine's cache branches, save_object(format = "qs2") without qs2, the
pathfinder method in the shared fit tests, and the remaining draws-format
values. One existing test is fixed: it passed variables = to
posterior::subset_draws(), which ignores it, so the "draws_array
without lp__" test had been running with lp__ all along.

Left uncovered on purpose: the WSL helpers (the WSL job runs the checks,
not covr, and adding covr there would take about as long again), code
that needs a failing CmdStan build or download, and guards in the
stdout state machine for output CmdStan doesn't print.

Copyright and Licensing

Please list the copyright holder for the work you are submitting
(this will be you or your assignee, such as a university or company):
Jonah Gabry

By submitting this pull request, the copyright holder is agreeing to
license the submitted work under the following licenses:

• Code: BSD 3-clause (https://opensource.org/licenses/BSD-3-Clause)
• Documentation: CC-BY 4.0 (https://creativecommons.org/licenses/by/4.0/)

[codecov-commenter]

Codecov Report

❌ Patch coverage is 88.00000% with 3 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (v1.0@f4bb79c). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
R/fit.R	72.72%	3 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff           @@
##             v1.0    #1291   +/-   ##
=======================================
  Coverage        ?   93.89%           
=======================================
  Files           ?       19           
  Lines           ?     7289           
  Branches        ?        0           
=======================================
  Hits            ?     6844           
  Misses          ?      445           
  Partials        ?        0           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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