feat(performance): token budget cap per run with abort

[bjridicodes]

Add max_tokens_per_run config setting. If a run would exceed it, abort early with partial results. Operational discipline: 'I keep token spend under 200k per run'.

[bjridicodes]

Relationships

Related to:

• feat(search): target specific job boards in LLM web search with per-board query cap #27 — both are cost control mechanisms; query cap controls LLM calls per run, budget cap is the backstop
• perf(tokens): reduce token consumption in LLM batch scoring #14 — token reduction strategies feed into the budget baseline