fix(crawler): compute page body hash once per page instead of twice

[YusukeHirao]

Summary

• --dedupe-cap's observation and update-page.ts's page_meta.body_hash write each independently called computeBodyHash on the same rendered HTML for every non-predicted internal page, since --dedupe-cap defaulted to enabled (Mark dedupe-cap trap pages post-hoc; default-enable --dedupe-cap #264). That synchronous hash (body extraction + two regex passes + SHA-256) ran twice per page on the single-threaded crawl event loop, compounding CPU pressure that can manifest as navigation timeouts and stalls under load.
• Generalizes the predicted-only precompute in Crawler#handleResult to run for every internal page with a rendered HTML body (mirroring update-page.ts's writeHtml && html.length > 0 write gate), and forwards the resulting hash through the page event → Archive.setPage → Database.updatePage → update-page.ts, which now prefers the precomputed value over recomputing it.
• The recompute path (bodyHash ?? computeBodyHash(page.html)) stays as a fallback for direct callers that have not precomputed one (e.g. existing test call sites), so this is purely additive at every layer's public signature.

Test plan

• yarn lint — 0 errors (pre-existing warnings only)
• NX_WORKSPACE_ROOT_PATH=<worktree> yarn build — all 13 projects
• yarn test — 4464 tests passing (3 new: page-event payload carries the precomputed hash; computeBodyHash called exactly once per internal page even with --dedupe-cap enabled; external pages never trigger the computation since setExternalPage never writes a body hash)
• yarn vitest run --config vitest.e2e.config.ts — full E2E suite (35 files / 166 tests), including dedupe-cap.e2e.ts's real crawl → archive → query pipeline
• Reviewed via /code-review medium, /qa-engineer, /product-manager — findings addressed (explicit bodyHash: null on the error-page emit path, unified @param bodyHash wording across archive.ts/database.ts/update-page.ts, added missing coverage for the external-page skip case)