test(e2e): consolidate cross-chain bridge and messaging suites

[spalladino]

Part of the e2e round-2 consolidation (PR 3 of 9): fewer CI containers for single-node/cross-chain/ with every asserted behavior preserved.

Merges (old → new)

• token_bridge_private.test.ts + token_bridge_public.parallel.test.ts + token_bridge_failure_cases.test.ts → token_bridge.test.ts: one CrossChainMessagingTest setup in beforeAll (with startProverNode: true, covering all three), with describe('private'), describe('public'), describe('failure cases'). The public file previously re-ran the full harness setup per it via beforeEach.
• l1_to_l2.parallel.test.ts → l1_to_l2.test.ts + l1_to_l2_inbox_drift.test.ts: beforeEach full setup converted to a single beforeAll per file; the scenarios converted to it.each(['private','public']) tables. Split by scenario (see timings below): l1_to_l2.test.ts keeps the duplicate-message-from-non-registered-portal scenario, l1_to_l2_inbox_drift.test.ts holds the inbox-drift scenario (which resets the markProvenEnabled gate between its via beforeEach).
• l2_to_l1.parallel.test.ts → l2_to_l1.test.ts: dropped the .parallel suffix; the two message-count-shape its ([3,4] and [3,1,2]) tableized via it.each with per-row consume tuples; the reorg-and-remine, mixed private/public, no-message-tx, and multi-block-checkpoint its kept explicit and behaviorally unchanged.

Harness construction

• The new CrossChainMessagingTest(...) for every suite now lives as the first statement of beforeAll rather than at describe scope, so nothing is constructed at test-registration time. Registration-time data (the it.each scope tables) is static, and test bodies read off t lazily.
• The shared L1→L2 message helpers (sendMessageToL2, advanceBlock, waitForMessageFetched, waitForMessageReady) used by both l1_to_l2 files are extracted to message_test_helpers.ts. Each suite plugs in its own proving policy via a markAsProven dependency (unconditional for the duplicate-message suite; gated for the inbox-drift suite).

Container count

• Before: 14 CI containers (2 plain files + 12 .parallel per-it containers), each paying a full node bootstrap.
• After: 4 containers / 4 setups. Same 17 its in total.

Timings (why l1_to_l2 was split)

Measured from the last full CI run on this branch (commit ba5e29b):

• token_bridge.test.ts: 463s (~7.7 min)
• l2_to_l1.test.ts: 384s (~6.4 min)
• l1_to_l2.test.ts: 738s (~12.3 min) — over the ~10 min target, so split

l1_to_l2's four its broke down as: duplicate-message private 149s / public 162s; inbox-drift private 194s / public 172s (setup + teardown ~59s). Splitting by scenario yields l1_to_l2 (~6 min) and l1_to_l2_inbox_drift (~7 min), both comfortably under 10 min, without resorting to the .parallel suffix.

Assertion mapping

• No assertions dropped. All 17 its and their asserted properties survive; every it moved intact to exactly one file.
• Because the token-bridge its now share one node, absolute balance assertions became delta-from-initial assertions (each it snapshots L1/L2 balances at its start). Same properties: deposit debits L1 by the bridge amount, claim credits L2, withdraw restores L1, wrong claims revert with the same error strings.
• .test_patterns.yml flake-pattern regexes repointed from the three old bridge file names to token_bridge; the l1_to_l2 prefix regex already covers both new l1_to_l2* files.
• bootstrap.sh: 25m per-container timeout overrides for the four cross-chain files (bodies run serially in one container each). The glob picks up the new file names unchanged.

Notes

• No docs #include_code references any of the renamed files, so no docs changes.
• single-node/README.md still lists the old cross-chain file names (it was already stale); left untouched to avoid conflicting with the sibling consolidation PRs — follow-up sweep will refresh it.