#8976: Add structured tracing for agent runs

This PR adds a low-overhead tracing layer for agent execution. When enabled via `--trace <path>`, agent runs record structured step data (LLM prompts, tool calls, outputs, and deterministic state hashes) to disk. The feature is disabled by default and does not alter execution behavior. Motivation: - Make agent failures reproducible - Improve debuggability without rerunning LLM calls - Provide a foundation for future replay and regression testing Scope: - Optional trace writer and schema - Deterministic state hashing - Minimal wiring at existing LLM/tool chokepoints - No replay, inspection UI, or behavioral changes All changes are additive and fully optional. <!-- greptile_comment --> <h2>Greptile Overview</h2> <h3>Greptile Summary</h3> This PR introduces an optional structured tracing subsystem for agent runs, enabled via `--trace <path>`. It adds a versioned trace schema (`TraceFile`/entries), deterministic state hashing, and a `TraceWriter` that persists LLM/tool step records plus final run metadata. The agent CLI plumbing wires `--trace` through `agent-via-gateway` → `agentCommand` → embedded runner params, and `runEmbeddedAttempt` records LLM calls and run completion when tracing is enabled. <h3>Confidence Score: 3/5</h3> - This PR is likely mergeable but has correctness/performance footguns in the tracing implementation. - Core wiring is straightforward and optional, but the current writer rewrites the full file on every flush (not append-only/low-overhead) and state hashing can throw depending on real message shapes, causing behavior changes when `--trace` is used. Schema parsing is also too permissive for a boundary function. - src/agents/tracing/writer.ts, src/agents/pi-embedded-runner/run/attempt.ts, src/agents/tracing/schema.ts <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} **Context used:** - Context from `dashboard` - CLAUDE.md ([source](https://app.greptile.com/review/custom-context?memory=fd949e91-5c3a-4ab5-90a1-cbe184fd6ce8)) - Context from `dashboard` - AGENTS.md ([source](https://app.greptile.com/review/custom-context?memory=0d0c8278-ef8e-4d6c-ab21-f5527e322f13)) <!-- /greptile_comment -->