refactor(prover): move inline table tests to tests/

[diegokingston]

Relocate the #[cfg(test)] mod tests blocks out of the table source files in prover/src/tables/ into prover/src/tests/, alongside the existing per-table test files.

• 7 new files: page/register/load/memw/memw_aligned/memw_register/ keccak_rnd _tests.rs
• decode and trace_builder inline tests appended to their existing decode_tests.rs / trace_builder_tests.rs
• is_register_op and collect_bitwise_from_keccak bumped to pub(crate) so the relocated trace_builder tests can reach them

bitwise.rs has no #[test] functions (only a #[cfg(test)] test helper) and is left unchanged; trace_builder's #[cfg(test)] impl helper methods also stay in place.

[github-actions]

Codex Code Review

No issues found in the PR diff.

The changes are primarily test relocation into prover/src/tests/*, plus two helpers widened to pub(crate) for those tests. I didn’t see security, correctness, or significant performance regressions in the changed code.

Verification note: I attempted cargo test -p prover --tests, but the sandbox could not run it because rustup tried to write to read-only /home/runner/.rustup/tmp; redirecting to /tmp then failed due blocked network/toolchain sync.

[claude]

Review: refactor(prover): move inline table tests to tests/

Clean refactoring — no security issues, no logic changes, no bugs introduced. Tests are mechanically moved, not modified.

Two minor points:

[Low] Redundant #[cfg(test)] on inner submodules in trace_builder_tests.rs
The keccak_tests and routing_tests submodules at lines 566/776 are wrapped in #[cfg(test)], but the entire prover/src/tests module is already gated by #[cfg(test)] in lib.rs (line 22). The inner wrappers are no-ops. Likely inherited from the original inline location where the guard was needed — safe to drop.

[Low] pub(crate) visibility bumps for is_register_op / collect_bitwise_from_keccak
These two functions are promoted from private to pub(crate) solely to support the relocated tests. The change is minimal and correct, but it means they're now callable from non-test code anywhere in the crate. If they're purely test-support helpers, consider leaving their signatures annotated with a comment to that effect — or alternatively, keep them private and use a #[cfg(test)] impl block (as is already done for other helpers in trace_builder.rs) to avoid widening their visibility at all.

[github-actions]

Codex Code Review

No issues found in the PR diff.

The production changes appear limited to moving tests out of implementation modules and widening a few helpers/constants to pub(crate) so crate-level tests can access them. I didn’t see security, VM correctness, significant performance, or simplicity problems.

I could not run cargo test because the sandbox has read-only /home/runner/.rustup/.cargo paths and restricted network access, so Cargo could not prepare dependencies.

[claude]

Review: refactor(prover): move inline table tests to tests/

Clean, low-risk refactoring. All tests are faithfully relocated with no logic changes or drops. A few minor observations:

Low – redundant #[cfg(test)] on nested modules

In trace_builder_tests.rs the new mod keccak_tests and mod routing_tests (lines 566 and 776) carry #[cfg(test)], but the enclosing file is already only compiled in test mode via #[cfg(test)] pub mod trace_builder_tests; in mod.rs. The attribute is a no-op. (Inline comment filed on line 566.)

Low – pub(crate) visibility bumps widen the internal API

Six items are now crate-visible that were previously module-private:

• select_storage_mode, GOLDILOCKS_BYTES, CUBIC_EXT_BYTES, KECCAK_NODE_BYTES in auto_storage.rs
• is_register_op, collect_bitwise_from_keccak in trace_builder.rs

These are implementation details needed only by the relocated tests. All stay behind pub(crate) (not pub), and auto_storage is already feature-gated, so this is an acceptable trade-off for the simpler refactor. Worth knowing if someone later reaches for these thinking they're part of an intentional internal interface.

Nothing else to flag

• The #[cfg(all(test, feature = "disk-spill"))] gate on auto_storage_tests is correct — auto_storage is itself behind #[cfg(feature = "disk-spill")] in lib.rs, so no test regression.
• INV_SHIFT_32 / SHIFT_32 are already pub const in templates.rs; templates_tests.rs compiles without further visibility changes.
• All five test_is_register_op_* routing tests are present in mod routing_tests at the end of trace_builder_tests.rs.