…t subscription drift (refs bsv-blockchain#872)

## Problem

Block assembly's listener goroutine sits idle indefinitely after a reorg
coincides with a blockchain-client gRPC EOF reconnect. Observed on
teratestnet: BA stuck at one height for 15+ hours while the blockchain
advanced 20+ blocks, only exposed by `WaitForBlockAssemblyReady` from
legacy/blockvalidation. Restarting the BA container clears the stall
because Start() runs a one-shot triggerReconcile().

The root cause sits in services/blockchain/Client.go subscription
lifecycle (likely a per-service ctx cancellation killing the subscriber
goroutine permanently — multiple Client instances per container, only
some share the long-lived daemon ctx). That investigation is out of
scope here.

## Fix

Defense-in-depth: add a ticker in `startChannelListeners` that calls
`triggerReconcile()` on a bounded cadence (default 30s). Each tick is
idempotent — if BA's tip matches the chain tip, processNewBlockAnnouncement
returns early with no work. If they differ (because notifications were
missed), the existing reconcile path runs and BA catches up.

Setting `blockassembly_periodicReconcileInterval`:
- Default 30s — recovers a silent subscription within 30s instead of
  15+ hours.
- 0 disables the ticker entirely (escape hatch for tests / dev).

New Prometheus counter `teranode_blockassembly_periodic_reconcile_total`
so operators can observe ticker firings (sustained high rate paired
with missing notification-driven reconciles signals subscription drift).

## Out of scope

- Root-cause fix in services/blockchain/Client.go subscription lifecycle.
  Needs instrumentation first (logging + counter on the
  line 1311 early-return) to identify which side fires (ctx.Err vs
  c.running) before changing semantics. Worth a follow-up issue.
- gRPC keepalive is already enabled and didn't catch this. Either the
  bug is hypothesis 1 (per-service ctx cancellation) or the half-open
  stream isn't actually what happened.

## Test plan

- New TestBlockAssembler_PeriodicReconcile_FiresWhenIntervalSet:
  50ms interval, asserts ≥2 ticker firings in a 220ms window past
  the startup baseline.
- New TestBlockAssembler_PeriodicReconcile_DisabledWhenIntervalZero:
  interval=0, asserts zero additional GetBestBlockHeader calls
  beyond startup.
- New TestBlockAssembler_PeriodicReconcile_NotificationStillProcessed:
  1h interval (effectively disabled within the test window),
  notification arrives on blockchainSubscriptionCh, asserts the
  notification path still drives a reconcile.
- `go test -race -count=1 -tags testtxmetacache ./services/blockassembly/`
  → 378 passed (no regressions).
- `go test -race -count=1 ./settings/` → 131 passed.
- `go vet ./services/blockassembly/ ./settings/` clean.
- `go build ./...` clean.

…kage convention

…tests

Address claude[bot] feedback on PR bsv-blockchain#874: the three new tests created
BlockAssembler instances but didn't explicitly tear down the listener
goroutine. BlockAssembler doesn't expose a Stop method (Stop lives on
the BlockAssembly server wrapper), so ctx cancellation is the supported
shutdown signal — see BlockAssembler.go:317.

Add a shutdownListenerGoroutine helper that cancels the test ctx in
t.Cleanup and gives the listener a 50ms grace window to exit. Pass the
cancellable ctx to utxostoresql.New, NewBlockAssembler, and ba.Start
so the entire chain shuts down deterministically.