ENH: Add Cragg-Donald test of reduced rank for first-stage regression (#622)

[AyuG619]

Summary

Implements the Cragg-Donald test of reduced rank for the first-stage
regression, addressing issue #622. With multiple endogenous variables,
the existing per-variable first-stage F-statistics can be individually
large even when the instruments cannot jointly identify all endogenous
regressors (e.g., when their fitted values are highly correlated). This
adds a joint diagnostic that correctly flags that failure mode.

Implementation

• New cragg_donald function in linearmodels.iv.common, alongside the
  existing f_statistic.
• New cragg_donald cached property on FirstStageResults, alongside
  the existing diagnostics property.
• Reuses the existing annihilate/proj utilities for partialling out
  exogenous regressors and projecting onto instruments, consistent with
  other diagnostics in this module.
• Returns a WaldTestStatistic/InvalidTestStatistic, matching existing
  conventions (e.g. sargan, wu_hausman, f_statistic).
• The reported p-value uses the asymptotic chi2(k-m+1) distribution
  (Anderson 1951), not Stock-Yogo (2005) finite-sample critical values,
  which require choosing a tolerance for maximal size distortion/bias
  and so are intentionally left out of scope.

Testing

• Validated against an independent reference implementation
  (mlondschien/ivmodels) on the exact example data from IV2SLS first_stage reports "wrong" first stage F-statistic #622, matching
  to ~10 significant figures.
• Degenerate-case tests for k < m (too few instruments) and m == 0 (no
  endogenous regressors) -- the latter caught by running the existing
  result_checker-based test suite in test_results.py, which exercises
  the OLS-like (no endogenous/instruments) fit path across all four IV
  estimator types.
• Full existing test suite (test_postestimation.py, test_results.py)
  passes with no regressions.

Closes #622

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.54%. Comparing base (273d7f1) to head (b214bc6).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #713   +/-   ##
=======================================
  Coverage   99.54%   99.54%           
=======================================
  Files         101      101           
  Lines       17426    17488   +62     
  Branches     1430     1432    +2     
=======================================
+ Hits        17347    17409   +62     
  Misses         29       29           
  Partials       50       50           

Flag	Coverage Δ
adder	99.53% <100.00%> (+<0.01%)	⬆️
subtractor	99.53% <100.00%> (+<0.01%)	⬆️

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

☔ 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.