Preserve tabled constants during module modernization

[prozacchiwawa]

Summary

• Remove the module modernization assignment that forced all constants to non-tabled.
• Collapse the remaining constant modernization branch into a guarded match arm to satisfy clippy.

Investigation notes

• The expected failures reproduce after preserving tabled constants:
  • cargo test test_export_foreign_constant -- --nocapture
  • cargo test test_program_exporting_constant_from_program -- --nocapture
• Temporary trace output around start_codegen, compute_env_shape, and nil-env mode showed the common phase keeps the tabled constant in the environment shape:
  • test-data.one.QUOTED_ONE: CommonPhase(false) env shape (test-data.one.QUOTED_ONE)
  • programs.single-constant.C: CommonPhase(false) env shape (programs.single-constant.C)
• In the following standalone pass, compile_module creates StandalonePhaseInfo with empty_common_phase: !common_phase_functions. For these constant-only modules, common_phase_functions is false, so empty_common_phase is true even though the common env is not empty.
• compute_env_shape then takes the StandalonePhase empty_common_phase branch. Since the tabled constant is already present in sp.env, it is excluded from extra_env_data, and the branch returns an empty left environment (()) for nil args. Nil-env mode subsequently rewrites (()) to ().
• Assertion check used during investigation: empty_common_phase should imply collect_env_names(sp.env).is_empty(). It failed with the live tabled constant still in sp.env, confirming the invariant violation.

Exact cause

empty_common_phase is derived solely from whether the common phase has defuns, not whether the common phase environment has live tabled constants. That was masked by d.tabled = false; once tabled constants are preserved, constant-only common phases can still have a non-empty environment, but standalone codegen treats it as empty and drops the tabled constants from the environment shape.

Verification

• cargo clippy --workspace -- -D warnings