#18997: fix: improve context overflow error messages and docs

## Summary Improves the user experience when hitting context overflow errors (#5771). ### Changes **Error messages (src)** - Include context window size, source, and message count in the overflow error so users can diagnose the issue - Add actionable suggestions: reduce workspace files, lower `bootstrapMaxChars`, use `/reset` - Improve diagnostic logging with `contextWindow` and `contextSource` fields **Documentation (docs)** - Document `agents.defaults.compaction` config with full example (mode, reserveTokensFloor, memoryFlush) - Document related settings: `contextTokens`, `bootstrapMaxChars`, `bootstrapTotalMaxChars` - Clarify that `autoCompact` is NOT a config key (common user confusion from the issue) - Add troubleshooting section for context overflow ### Before ``` Context overflow: prompt too large for the model. Try again with less input or a larger-context model. ``` ### After ``` Context overflow: prompt too large for the model. Context window: 200,000 tokens (source: default). Messages in session: 3. Try /reset (or /new) to start a fresh session, reduce workspace files (AGENTS.md, SOUL.md, etc.), lower `agents.defaults.bootstrapMaxChars`, or switch to a larger-context model. ``` ### Testing - [x] `pnpm build` passes - [x] Existing unit tests pass (`run.overflow-compaction.test.ts`, `config.compaction-settings.test.ts`) - [x] AI-assisted (Claude via OpenClaw) Closes #5771 <!-- greptile_comment --> <h3>Greptile Summary</h3> Improves context overflow error messages with actionable diagnostic information (context window size, source, message count) and adds comprehensive compaction documentation including config examples, related settings, and a troubleshooting section. - `run.ts`: Error message now includes dynamic context info (window size, source, message count) and specific remediation steps - `agent-runner-execution.ts`: Catch-all fallback message updated with consistent suggestions (/reset, reduce workspace files, switch model) - `compaction.md`: Added YAML config example, related settings table, `autoCompact` clarification, and troubleshooting section — all verified against source code - **Issue found**: The e2e test in `reply.triggers.trigger-handling.ignores-inline-elevated-directive-unapproved-sender.e2e.test.ts` (line 140-142) uses `toBe()` exact match against the old fallback message from `agent-runner-execution.ts` and was not updated in this PR — this test will fail <h3>Confidence Score: 3/5</h3> - Low-risk changes to error messages and docs, but an existing e2e test will break due to an exact string match on the old message. - The code changes are straightforward improvements to error messages and logging, and the documentation additions are accurate. However, the e2e test `reply.triggers.trigger-handling.ignores-inline-elevated-directive-unapproved-sender.e2e.test.ts` expects the old message string via `toBe()` and was not updated, which will cause a test failure. Score of 3 reflects that the functional changes are sound but the PR is not merge-ready until the test is fixed. - Pay close attention to `src/auto-reply/reply/agent-runner-execution.ts` — the updated error message at line 544 breaks an e2e test that was not included in this PR's changes. _{Last reviewed commit: 1eef057} <!-- greptile_other_comments_section --> _{(2/5) Greptile learns from your feedback when you react with thumbs up/down!} <!-- /greptile_comment -->