Allow external adjustment of parallel search workers

[zayenz]

Parallel portfolios sometimes need to move a fixed CPU budget between search
engines while those engines keep running. This adds Search::WorkerControl, a
copyable external handle for changing how many of a parallel engine's
pre-created workers may execute.

The initial threads value remains the fixed resident capacity. A control
starts with a request between one and that capacity; later request() calls
are thread-safe and non-blocking. Growing wakes parked workers. Shrinking is
cooperative: excess workers finish their current search action before parking.

The control applies to parallel DFS and BAB, including leaf engines owned by
RBS and PBS. PBS deliberately keeps one control per asset; allocation policy
and any portfolio-wide budget invariant remain with the caller.

The handle binds to one logical leaf-engine lifetime. Copies share the same
request state, and an empty handle preserves existing fixed-worker behavior.

Validation includes:

• a release build of the complete gecode-test binary;
• the existing randomized DFS and BAB matrix, with mixed fixed and adjustable
  engines plus sampled asynchronous resizing;
• the existing RBS and PBS matrices, including independently adjusted explicit
  portfolio assets; and
• a no-thread build plus focused worker-control API and lifecycle checks.