Add bounded capture concurrency

[pudo]

Pravda must limit concurrent browser captures so incoming traffic cannot exhaust browser, memory, or CPU capacity.

Add per-instance admission control around snapshot creation.

Behavior

• Configure a maximum number of concurrent captures.
• Queue excess requests up to a bounded limit.
• Reject requests when the queue is full.
• Apply a maximum queue wait time.
• Release capacity when a capture succeeds, fails, times out, or is cancelled.
• Expose active captures and queue depth through structured logs and metrics.

Questions

• Should queued requests remain open or receive 429 immediately?
• Should concurrency be limited per Pravda instance, browser process, or both?
• How many concurrent contexts can one playwright run-server process support reliably?
• Should limits be fixed configuration or derived from browser-worker capacity?
• Does the deployment topology selected in Define the Pravda Kubernetes deployment model #7 require different admission behavior?

Acceptance criteria

• Concurrent captures never exceed the configured limit.
• Queue depth and wait time are bounded.
• Saturation produces a documented HTTP response.
• Failed and cancelled captures release capacity.
• Tests cover execution limits, queue saturation, timeout, and cancellation.
• A small load test establishes an initial concurrency setting for the deployment model.