refactor(executor): move inline unit tests to src/tests/ module

[diegokingston]

Summary

Relocates the three inline #[cfg(test)] mod tests blocks in executor/src into a dedicated executor/src/tests/ module, matching the convention already used by the prover and math crates.

From	To	Covers
flamegraph.rs	tests/flamegraph_tests.rs	demangle
vm/instruction/execution.rs	tests/keccak_tests.rs	keccak-f[1600] + Keccak syscall
vm/memory.rs	tests/memory_tests.rs	public-output commits, load bounds

demangle and Memory::set_bytes_aligned were module-private; bumped to pub(crate) so the relocated tests can reach them. No test logic changed — bodies moved verbatim, only use super::* rewritten to explicit crate:: paths.

Test plan

• cargo clippy -p executor --all-targets -- -D warnings — clean
• cargo test -p executor --lib — 12 passed (1 demangle + 4 keccak + 7 memory)

[github-actions]

Codex Code Review

No security vulnerabilities, behavioral bugs, or significant performance issues found in the PR diff.

One minor cleanliness note: executor/src/lib.rs now exports pub mod tests; in normal builds, but the submodules are #[cfg(test)], so it becomes an empty public executor::tests module outside tests. That is likely harmless, but #[cfg(test)] mod tests; would avoid exposing test-only API surface.

I attempted cargo test -p executor --no-run, but the environment failed before compiling because rustup could not create a temp file under the read-only /home/runner/.rustup/tmp.

[claude]

Review

Clean refactor overall — test bodies are moved verbatim and the logic is unchanged. One bug to fix before merging, plus a minor nit.

Bug — missing #[cfg(test)] on the tests module (Medium)

executor/src/lib.rs declares pub mod tests; without a #[cfg(test)] gate. The two crates cited as the convention to match (prover, math) both use:

#[cfg(test)]
pub mod tests;

Without the gate the tests module ships into every non-test build and becomes part of the crate's public API (executor::tests). It will be empty in those builds (all sub-modules in mod.rs are individually gated), but it still pollutes the public surface and diverges from the stated goal.

Nit — pub(crate) on set_bytes_aligned (Low)

The function's own doc comment says it "Should only be used to write to public output and private input." pub(crate) makes it reachable from anywhere in the crate, widening the accidental-misuse surface. Worth keeping it narrower if possible (e.g. pub(super) or a test-only re-export).

[github-actions]

Benchmark Results for unmodified programs 🚀

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base binary_search	50.6 ± 2.5	48.8	57.4	1.00
head binary_search	52.0 ± 5.2	48.6	65.3	1.03 ± 0.12

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base bitwise_ops	50.1 ± 0.8	48.7	50.9	1.00 ± 0.03
head bitwise_ops	50.0 ± 1.4	48.5	52.2	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base fibonacci_26	55.2 ± 0.9	54.1	56.5	1.00
head fibonacci_26	56.5 ± 3.2	53.9	64.2	1.02 ± 0.06

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base hashmap	106.5 ± 1.7	104.6	109.5	1.00
head hashmap	107.7 ± 3.0	103.5	114.8	1.01 ± 0.03

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base keccak	108.5 ± 0.7	107.5	109.6	1.00 ± 0.02
head keccak	108.4 ± 1.7	105.6	110.7	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base matrix_multiply	61.9 ± 1.1	60.4	63.9	1.01 ± 0.03
head matrix_multiply	61.6 ± 1.6	59.9	65.4	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base modular_exp	50.5 ± 2.1	49.0	56.0	1.00
head modular_exp	50.6 ± 1.0	49.1	51.8	1.00 ± 0.05

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base quicksort	53.2 ± 0.7	51.7	54.0	1.01 ± 0.02
head quicksort	52.5 ± 0.8	51.6	53.9	1.00

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base sieve	55.9 ± 1.1	54.7	57.6	1.00
head sieve	55.9 ± 1.1	54.6	57.4	1.00 ± 0.03

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base sum_array	66.5 ± 1.0	65.0	67.9	1.00
head sum_array	67.0 ± 1.2	65.7	68.8	1.01 ± 0.02

Command	Mean [ms]	Min [ms]	Max [ms]	Relative
base syscall_commit	78.6 ± 2.8	76.0	85.2	1.00
head syscall_commit	82.8 ± 9.4	75.4	102.4	1.05 ± 0.12