Cap clock-resolution estimation iterations to avoid OOM

[22elix3r]

Summary

Clock-resolution estimation in Catch2's benchmark environment probe is only bounded by a hard-coded 500ms window. On a cheap, high-resolution clock (for example an embedded qemu-system-arm target with 16MB RAM) that window can request well over a million TimePoint samples, exhausting memory inside resolution() and throwing std::bad_alloc.

This change adds a sample cap so warmup() and estimate_clock_resolution() stop doubling once 100000 iterations are reached, even if the time budget has not elapsed.

Related Issue

Fixes #3180

Changes Made

• Added clock_resolution_estimation_iteration_limit (100000) next to the other environment-probe constants.
• Extended run_for_at_least with an optional max_iterations argument (default remains the historical 1 << 30 "optimized away" ceiling).
• Passed the new cap from warmup() and estimate_clock_resolution(), which are the two paths that allocate O(n) TimePoints via resolution().
• Left the two-phase sampling in resolution() unchanged so consecutive Clock::now() calls are still measured with minimal work between them.
• Did not add a new CLI flag: the OOM is caused by unbounded sample count, not the time budget, and the environment probe is cached once per process. A hard cap is the safer default for memory-constrained targets.

Testing

• Added clock resolution estimation is iteration-capped using counting_clock at 1ns/tick, which would otherwise demand hundreds of millions of samples to fill 500ms.
• Added run_for_at_least respects max_iterations.
• Extended estimate_clock_resolution to assert samples_seen stays within the cap.
• Updated the run benchmark fake-clock tick count for the capped probe.
• Regenerated SelfTest approval baselines for the new cases.
• ctest in a Debug basic-tests build: 82/82 passed, including RunTests and ApprovalTests.

Notes

• Adding safeguard to clock resolution estimation (Issue 3180) #3182 also targets this issue. That patch rewrote estimate_clock_resolution and stopped storing every TimePoint, but it did not cap warmup() (which uses the same resolution() allocator), left SelfTest's exact fake-clock assertion failing, and has had no maintainer review. This PR keeps run_for_at_least as the single doubling loop, caps both warmup and resolution collection, and updates the tests.
• 100000 matches clock_cost_estimation_tick_limit already in the same header. Peak extra allocation is about 1.6MB (TimePoints + deltas) instead of ~20MB at the 1.28 million samples reported in Clock resolution estimation iterations limit #3180.
• Amalgamated distribution files were not updated, per contributing guidelines.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.18%. Comparing base (317ac1e) to head (8e2256a).

Additional details and impacted files

@@            Coverage Diff             @@
##            devel    #3196      +/-   ##
==========================================
- Coverage   91.21%   91.18%   -0.03%     
==========================================
  Files         206      206              
  Lines        9029     9033       +4     
==========================================
+ Hits         8235     8236       +1     
- Misses        794      797       +3     

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[22elix3r]

Ready for review. I cannot request a reviewer from this account.

[22elix3r]

Withdrawing: I missed that this issue already had earlier open PRs. Sorry for the noise — leaving the field to those patches.

[22elix3r]

Withdrawing this PR so the earlier open work on the issue can proceed.