refactor(core): centralize full-prefix KV scans - #45583

Why

Background-job recovery currently has to manage KV pagination itself, mixing storage cursor handling with recovery-record decoding. It also uses the default 100-entry page size even though it needs every matching record.

What Changes

Add KV.scanAll(prefix) to collect matching entries in key order using the existing paginated scan with 1,000-entry pages. Job.pendingBackground uses it and keeps the same recovery-record decoding and invalid-record filtering.

The existing scan limits and cursor contract remain unchanged. Focused assertions cover literal special-character prefixes, ordering, pagination, and repeated evaluation.

Verification

# Repository root
bun install --frozen-lockfile
bunx prettier --check packages/core/src/kv.ts packages/core/src/job.ts packages/core/test/kv.test.ts
git diff --check
git diff --cached --check
git push -u origin kv-scan-all

# packages/core
bun typecheck
bun run test test/kv.test.ts test/job.test.ts
bun run test test/session-execution.test.ts test/tool-shell.test.ts

• Dependency installation, formatting, whitespace checks, and core typechecking passed.
• Focused KV/job tests: 15 passed, 0 failed, 147 assertions.
• Recovery and shell tests: 98 passed, 19 skipped, 0 failed, 552 assertions. The skipped tests require PowerShell, which is unavailable locally.
• The normal push hook ran the repository typecheck: 33 successful tasks, including 27 cache hits. Hooks were not bypassed.
• The full core test suite and live application workflows were not run.