chore(runtime): retire the test-only V1 HistoryCompactBlock path and Synthesis Cache

[Astro-Han]

Found by a repo-wide simplification audit; re-verified with git grep -aI. Independently confirmed by two separate audit passes.

The V1 HistoryCompactBlock artifact compaction path and the Synthesis Cache are test-only:

• The sole production composer of AiSdkBackend (packages/runtime-host/src/server/execution-model-composition.ts, checkpoint wiring around :385) supplies only loadHistoryCompactCheckpoint / summarizeHistoryCompact / recordHistoryCompactCheckpoint; it never supplies loadHistoryCompact / writeHistoryCompact.
• All 9 repo-wide producers of writeHistoryCompact: are in packages/runtime/src/__tests__/ai-sdk-backend.test.ts.
• history-compact-artifacts.ts and synthesis-cache-artifacts.ts have no production callers (tests + package export only).
• MAKA_CONTEXT_SYNTHESIS_CACHE=on is exposed (context-budget-policy.ts) but there is no production persistence capability behind it — a toggle that cannot do anything.
• docs/architecture/llm-compaction-events-log-projection-draft.md still describes V1 artifacts as a read-compat path, contradicting the actual wiring.
• The only real capability user is the benchmark script scripts/deepseek-live-cost-baseline.mjs (itself orphaned; see the audit-residue issue).

Suggested cleanup: delete the two artifact adapters + tests + exports; move or migrate the benchmark's legacy usage; drop the unreachable branches and the false toggle; fix the doc. If shipped artifacts/local DBs may still hold V1 blocks, a minimal decoder can be retained — but not the full write path.

[likun666661]

I'd like to claim this cleanup. I worked on this compaction area previously, so I can also verify the historical intent and compatibility boundaries rather than treating the current zero-consumer state as the whole story.

I plan to build on #3242 once its audit-residue cleanup lands, keep the current ledger/checkpoint compaction path intact, and remove the unreachable V1 artifact and Synthesis Cache capabilities, adapters, configuration, tests, and stale documentation. Before deleting any compatibility surface, I'll verify the old production wiring and whether a minimal decoder or explicit migration note is still warranted.

This claim note was drafted with Codex assistance; I will review and take responsibility for the implementation and submission.