test(substrate): expand coverage via mock.module — claim/complete/fail/lease-recovery paths

Closes the test-gap finding from arc0btc #4 and secret-mars: prior
suite covered only `resolveSubstrateConfig`, `jobRowToTaskInput`, and
two write-back *early-return* guards via a fakeDb that threw if
touched. Claim, complete, fail, epoch-mismatch no-op, transient
write-back throw, and lease recovery — the "highest correctness
surface" per the PR's own Commit-2 description — were untested.

Rewires the file to use `mock.module("@genesis-works/substrate-db", ...)`
with mock fns per substrate-db export. Local DB is a real
`openDb(:memory:)` so `enqueueTask` exercises the actual sqlite write
path. console.{info,error,warn} are spied per-test for log-line
assertions.

New tests (18, alongside 10 existing for 28 total):

- `runSubstrateIntakeTick`:
  - happy claim → enqueues local task, returns `claimed=true` with
    correct epoch + jobId, emits `[substrate] claim ...` log
  - null claim → silent (no `[substrate] claim` log line; just
    `claimed: false`)
  - db-unreachable → catches `claimNextJob` throw, logs
    `reason=db-unreachable`
  - local-enqueue-fail → forces `enqueueTask` to throw by closing the
    local db, asserts catch fires + log line includes
    `reason=local-enqueue-fail jobs.id=<id>` (proves Copilot #4 /
    arc0btc #1 fix is wired)

- `runSubstrateWriteBack`:
  - non-substrate source no-op (existing — kept)
  - missing `_substrate_job_id` no-op (existing — kept)
  - complete happy → calls `completeJob` with `claim_epoch` arg,
    `[substrate] complete jobs.id=... epoch=...` log
  - complete `{ ok: false }` → emits self-contained fallback
    `[substrate] complete-failed` log (proves secret-mars #5 fix)
  - complete throws → catches transient PG blip, emits
    `[substrate] write-back error=... jobs.id=...` (proves arc0btc #3
    fix is wired — write-back no longer propagates out of `runOnce`)
  - fail happy → calls `failJob`, logs reason snippet
  - fail `{ ok: false }` → emits `[substrate] fail-failed` fallback

- `runSubstrateLeaseRecovery`:
  - released > 0 → logs `[substrate] lease-recovery released=<n>`
  - released = 0 → no log (quiet success)
  - throws → catches, logs `[substrate] lease-recovery error=...`,
    does NOT propagate

- `createSubstrateConnection`:
  - throws when `substrate.host` is unset (proves Copilot #3 fix —
    no implicit default to a hard-coded private IP)
  - throws on whitespace-only host

- `jobRowToTaskInput`:
  - asserts `payload.idempotency_key = "substrate-<job_id>-e<epoch>"`
    (proves idempotency-key threading is wired)
  - asserts `priority = 1`
  - asserts `_substrate_*` payload fields threaded
  - asserts `max_attempts = 1` (substrate handles retry)
  - existing source / subject-fallback tests kept

Header comment rewritten to match what's actually covered (Copilot #7).
Unused `mock`/`beforeEach`/`afterEach` imports replaced with the ones
actually used (Copilot #6 — though now they ARE used, so the prior nit
naturally resolves).

Test count: 143 total (28 substrate + 115 existing) — up from 125.
`bunx tsc --noEmit` clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: whoabuddy