Restore the process runtime state the money journeys pin

[Parad0x-Labs]

What failed

Shard tests (3) of CI run 35640946053 (PR #22) failed four cases with
cryptography.exceptions.InvalidTag at network/signer.py:302:

• tests/test_identity_lifecycle.py::test_mesh_message_rejects_revoked_peer
• tests/test_identity_lifecycle.py::test_signed_write_rejects_revoked_peer
• tests/legacy/test_learning.py::test_swarm_learning
• tests/legacy/test_pow_genesis.py::test_sybil_defense

These four also fail on main (8153a96) when the same files run in one
process in this order — reproduced locally in both trees:

pytest tests/effect_budget/test_money_served_api.py::test_the_daemon_serves_money_state_written_by_independent_processes_and_keeps_it_across_restart \
       tests/test_identity_lifecycle.py tests/legacy/test_learning.py tests/legacy/test_pow_genesis.py
→ 4 failed (InvalidTag) on main AND on PR #22's head; 9 passed with this PR.

Main's CI never showed it only because the round-robin shard split (descending
file size) puts test_money_served_api in shard 3 and test_identity_lifecycle
in shard 2 — PR #22's new guard file shifted the indices and collided them.
Any future test-file addition can reshuffle shards the same way.

Root cause

money_race_probe.configure_home pins FOUR pieces of process-global state
(VOOL_HOME env, runtime-home override, default db override, continuity db
override), but every test-side restore was partial:

• the served-daemon journey's finally restored the override + continuity
  path — never the VOOL_HOME env write (active_vool_home() reads env
  first, so the drift survives configure_runtime_home(None));
• shared_home had the same two-line partial restore;
• both provider_home fixtures "restored" with configure_runtime_home(None),
  which CLEARS the session home the root conftest installed instead of undoing
  their pin.

The race children (VOOL_KEY_PASSPHRASE="money-race-probe") seal their node
signer record under the pinned served home; after the journey, the next
signer reload in the same pytest process re-resolves _KEY_DIR into that home
and opens the record with the suite passphrase
(test-suite-only-never-a-real-key) → InvalidTag.

Repair (test-infrastructure only; no production semantics touched)

• money_race_probe.runtime_state_snapshot() / restore_runtime_state()
  snapshot and put back exactly the four pinned pieces, beside
  configure_home (the mutation owner); pinned_runtime() composes them for
  context-manager callers. The CLI children keep plain configure_home
  (the mutation dies with their process).
• The served-daemon journey, the cross-process suite's shared_home, and both
  provider_home fixtures restore exactly what they pinned.
• New guard case test_the_served_daemon_test_leaves_the_process_runtime_home_behind_it
  runs directly after the served-daemon journey in every shard/ordering (same
  module ⇒ same shard ⇒ same order) and pins the invariant against the
  module-import snapshot. Verified it fails with the drift message when the
  restore is removed, and passes with it.

Validation

• Poisoned four-file order: 4 failed on main → 9 passed with this PR (A/B via
  stash on the same tree).
• tests/effect_budget/ package: 174 passed, 1 failed —
  test_door_command_machine.py::TestZeroBudgetRefusal::test_zero_command_budget_means_zero_subprocess_transport_and_journal_calls,
  which fails identically on clean main locally and in every CI census (main
  35570948370 and PR runs 35637092451/35637818825/35640946053/35642039754) —
  pre-existing baseline, untouched here (manual-mode approval genus).
• Cumulative neighbors: effect_budget + daemon_message_security +
  network_identity_defects + maintenance_dht_discovery + mesh_workproof_reward
  • capability_tokens: 211 passed, 3 xpassed, same single baseline failure.
• Signer families in one process: money served + money cross-process +
  identity_lifecycle + legacy learning/pow + signer_key_storage +
  secret_storage_truth: 42 passed.
• ruff check . clean (pinned 0.16.7).

Base: main 8153a96. Head: e904c27. Test-only diff (5 files under
tests/effect_budget/).