Fetch complete history for pytest reference checks

[luisleo526]

The pytest checkouts are shallow, so historical code-generation comparisons cannot read their reference commits. Fetch complete branch history in both pytest jobs while preserving action pins, permissions, credential handling and event guards.

This is the CI-only prerequisite for a separate two-test reference repair. Compiler, gate and test source bytes are unchanged. A full-history control on accepted main paired with the current engine recorded 8,101 passes, two pre-existing whole-C++ comparison failures and 12 skips; corpus, differential gate and package checks passed. The failing cases are test_constant_and_input_lengths_keep_their_cpp and test_scripts_without_a_read_keep_their_cpp for the generated-history-member name. Control receipt SHA-256: bd570c7bc48c3b198db38cd16eb1c73aab5056db4115ec9c8c09e9dc107b648c. The final CI rollup remains required before landing. No other failed check is accepted by the sequencing decision.

The reference repair follows after this prerequisite lands and promotes. Detached historical objects remain a separate verification prerequisite; complete branch history does not prove they are available. The Test workflow runs on main pushes or manual dispatch, while PR gating uses the trusted base workflow. Actual CI outcomes will be recorded rather than predicted.

Validation: the complete sweep measured all 7,989 probes across 134 cases. Published snapshot ea0ef12764db6fa48c7bc13460489b0dbce0311b4a2dc98e08111b4ae88d8d66 matches the accepted baseline on every grade row and coverage field. The measured gate returned PASS_NO_IMPROVE_NO_REGRESSION with zero failures, hard moves, tier changes or coverage gaps. Both independent exact-head source reviews returned MERGE. GitHub check results remain pending and will be read in full before landing.