fix(core): drop the third-identical-call veto from the turn loop

[Max17190]

Why

The turn loop refused the third consecutive tool call with identical arguments and told the model "the result will not change." Arguments say nothing about results: a wait on a delegated child that times out at the bash cap, a git status while a build runs, a log tail all repeat verbatim while their output moves. The delegate skill's openmax-tmux wait needs exactly this shape for any child running past ten minutes (300 s bash cap, three identical waits), and the third was vetoed with a false message. The veto was added for small local models that loop; a stuck model is already bounded by max_agent_iterations and max_agent_tokens, which are budgets rather than a guess about the model.

Summary

• Remove RepeatCallTracker and its plumbing from the serial and batch dispatch paths (agent.rs).
• Collapse the now single-statement if executed { if mutating { into one condition; the large hunk in the diff is that re-indent (git diff -w shows the real change).

Test Plan

• New identical_consecutive_calls_all_execute: a scripted endpoint returns the same bash call on every request; with max_agent_iterations = 4 all four calls execute and the turn ends at the iteration cap. Fails on the previous loop (calls three and four returned the veto text).
• Reproduced the veto on the HEAD build against a mock endpoint before the change: three identical bash: date +%s%N; echo poll calls returned different timestamps and the third was refused.
• cargo test --workspace green; cargo clippy --workspace --all-targets at zero warnings.

Greptile Summary

This change removes the third-identical-tool-call veto so polling operations can continue when their arguments remain stable. The repeated-call flow was exercised against a scripted endpoint: four identical bash calls completed successfully, and the configured iteration limit ended the turn after the fourth call. The concern that removing the veto could allow repeated calls to continue past the configured limit was disproved by this execution.

Confidence Score: 5/5

The change is safe to merge based on the exercised repeated-call and termination behavior.

The focused regression execution covered repeated identical calls, successful tool results, endpoint request counting, and termination at the configured iteration limit. The same test distinguishes the previous behavior, where later calls were rejected by the removed veto.

Files Needing Attention: No additional files need attention; the reviewed behavior is contained in crates/core/src/agent.rs.

T-Rex Logs

What T-Rex did

• Ran cargo test -p open-max-core identical_consecutive_calls_all_execute --locked on the PR revision; the test completed with one passing test, four identical poll calls, and termination at max_iterations.
• Compared the same focused test against the parent revision; the parent version vetoed calls 3 and 4 with the identical-call message, showing a regression difference from the PR.
• Documented the exact reproduction workflow and results, including the before-PR reproduction script and log, and the after-PR focused-run script and log, which show the PR passes.
• Linked and reviewed artifacts that support validation, including the scripts and logs for both the parent and PR revisions.

_{Ran code and verified through T-Rex}

_{Reviews (1): Last reviewed commit: "fix(core): drop the third-identical-call..." | Re-trigger Greptile}