feat: salvage malformed compress content strings (omp#121)

[ranxianglei]

Why

Weak/local models (vLLM qwen etc.) emit compress tool arguments that fail strict JSON.parse ~50% of the time — truncated output caps, raw newlines inside summary strings, trailing commas, or plain prose (see billion-context-omp#121). The ACP host extension's normalizeRanges threw Invalid content: not valid JSON on the first strict-parse failure and relied entirely on the retry nudge — but weak models fail the retry the same way ~50% of the time, so compress became effectively unusable with local models.

What

Before throwing, run the kernel's salvageParseRanges ladder (fences / comma+newline repairs / truncated-array prefix / field-regex): recoverable entries proceed as normal ranges (logWarn records the layer); truly unparseable input still throws, so the isError:true → retry-nudge contract is unchanged.

Bumps acp-kernel to ^0.0.34 (introduces salvageParseRanges + extractRanges double-encode normalization).

Tests

New truncated-content salvage case in tests/compress-retry.test.ts (first complete entry recovered from a blob truncated mid-second-entry). Full suite 410/410, typecheck clean.

Companion PRs: acp-kernel#109 (parser), billion-context (proxy), billion-context-omp (replay).

--- DEPENDENCY NOTE (pre-publish) ---

⚠️ Do not merge yet: package.json currently pins acp-kernel to a local file: tarball so the diff is reviewable now. Depends on acp-kernel PR #109 landing and npm publish of 0.0.34 (0.0.33's publish failed with npm 404 — expired NPM_TOKEN). Before merge this PR will be updated to "acp-kernel": "^0.0.34" from from the registry and the full suite re-run.