[1.0] Enforce context and execution budgets and show available usage

Default context can grow with the whole visible blackboard, while engine limits and usage reporting are uneven. Streaming fragment counts cannot be treated as token accounting.

Scope

• Use existing context providers with explicit size/selection limits and observable budget decisions; do not silently change shared-blackboard semantics.
• Make run time and supported tool-call limits effective along the qualified execution path.
• Associate available provider token/cost values with the run and preserve unknown values when providers do not report them.

Acceptance criteria

• An oversized visible history stays within the configured context budget, with omitted/truncated context made explicit and visibility still enforced.
• Controlled timeout/tool-call-limit cases end truthfully and release execution capacity.
• Missing usage appears as unknown, not zero; streaming chunks are not reported as token usage.

Boundaries

Cross-provider billing normalization and sophisticated monetary budgets are deferred.

References

• Implementation dependencies: #440.
• src/flock/core/context_provider.py
• src/flock/engines/dspy_engine.py
• src/flock/orchestrator/context_builder.py

Start with src/flock/core/context_provider.py and src/flock/orchestrator/context_builder.py to trace context selection and budget decisions, then follow the qualified execution path in src/flock/engines/dspy_engine.py. Check dependency issue #440 before changing behavior. Done means oversized context is explicit and bounded, timeout and tool-call limits release capacity, and missing provider usage remains unknown rather than zero.