Add worker crash recovery to ProcessParallelController

[jhewers-pf]

When a child process in the ProcessPoolExecutor crashes (OOM, segfault, etc.), Python raises BrokenExecutor and the pool becomes unusable. Previously, this was caught as a generic Exception, logged, and caused silent failure of the evolution process.

Changes

• Add explicit BrokenExecutor exception handling in run_evolution()
• Add _recover_process_pool() method that gracefully shuts down the broken executor, runs garbage
  collection, waits briefly for system stabilization, and recreates the pool
• Re-queue all pending iterations after recovery
• Track recovery attempts with a limit of 3 consecutive failures to prevent infinite loops
• Reset recovery counter after successful iterations (only consecutive crashes count toward the limit)
• Propagate BrokenExecutor from _submit_iteration() for centralized handling

Behavior
When a worker crashes:

1. Detect BrokenExecutor exception
2. Collect all pending iteration numbers
3. Shut down broken pool, run GC, wait 2s
4. Recreate fresh pool
5. Re-queue failed iterations
6. Continue evolution

If 3 crashes occur without any successful iterations in between, evolution stops gracefully.

[CLAassistant]

All committers have signed the CLA.