Parallel CO-BRAVO PoC

[Shnatsel]

Same approach as #97 but applies to CO-BRAVO tiles instead of the inner BRAVO

While the first two commits are 100% safe, the third introduced unsafe just to see what the performance ceiling is without any overhead. Still slower than sequential on 2 threads on my machine with dual-channel memory; needs measurement on servers with more memory channels.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.91%. Comparing base (3aa61ad) to head (2a16026).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #107      +/-   ##
==========================================
- Coverage   98.87%   97.91%   -0.97%     
==========================================
  Files          10       10              
  Lines        2227     2253      +26     
==========================================
+ Hits         2202     2206       +4     
- Misses         25       47      +22     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[Shnatsel]

No change on machines with dual-channel memory (presumably 2 threads already saturate memory bandwidth) but on an 8-channel server this gives a 30% improvement end-to-end, although the safe version only achieves a 10% improvement. We need to bring down the overhead of tracking disjoint regions, and add heuristics for selecting the parallel impl over the default one.