test(e2e): name shared timing presets and collapse proof-submission constants

[spalladino]

Round-2 e2e consolidation, PR 5 of 9 (workstream B: timing/config constants). Introduces two named timing constants in single-node/setup.ts and sweeps single-node/multi-node files onto them.

Constants added

• NO_REORG_SUBMISSION_EPOCHS = 1024 — proof-submission window so large the chain never prunes/reorgs in the test's lifetime. Re-exported via multi_node_test_context.ts and the per-directory setup re-exports.
• PROVING_SLOT_TIMING = { ethereumSlotDuration: 4, aztecSlotDurationInL1Slots: 3 } — the "12s floor" cadence previously hand-spelled (with near-verbatim derivation comments) in 5 proving/partial-proofs files. One shared doc comment now carries the derivation; the per-file copies are deleted.

Pure renames (same effective value)

• aztecProofSubmissionEpochs: 1024 → constant: setupBlockProducer, MOCK_GOSSIP_MULTI_VALIDATOR_OPTS, missed_l1_slot, multi_validator_node.parallel.
• PROVING_SLOT_TIMING adoption with unchanged values: single_root, multi_proof, long_proving_time, upload_failed_proof.

Nominal value changes, same "never reorg" semantic (large window → 1024)

• 640 → constant: bot, fees/failures, fees/fee_settings, block-building/debug_trace.
• 1000 → constant: proving/long_proving_time, proving/optimistic.parallel (6 sites).
• 128 → constant: sequencer/gov_proposal.parallel.

These windows only exist to pin blocks against pruning; none of the tests run anywhere near the window, so 1024 is inert. Verified by build/lint plus sample runs (below).

Effective timing changes (each run locally twice, both green)

• 16s-cluster collapse onto PROVING_SLOT_TIMING (16s → 12s aztec slot): proving/proof_fails.parallel (218s/218s), recovery/prune_when_cannot_build (83s/83s).
• Epoch one-off audit: partial-proofs/multi_root epoch 1000 → 32 (232s/233s), recovery/manual_rollback epoch 100 → 32 + PROVING_SLOT_TIMING + never-reorg window (58s/58s).

No fallbacks were needed; all collapses held over both runs. Rename-only sample runs: single_root 66s, long_proving_time 191s, upload_failed_proof 51s — all green.

Deliberately left as literals

• Small windows that are production-like or test the window itself: 1 (prune_when_cannot_build, l1-reorgs/setup, sync/synching, proof_boundary.parallel), 2 (cross-chain/l1_to_l2.parallel), 15 (escape_hatch_vote_only, needed for a valid EscapeHatch config — already commented).
• single-node/l1-reorgs/** values untouched per round-1 finding that their proof windows are sensitive to ethereumSlotDuration.

Skipped (sibling-PR file ownership; literal sweeps deferred to a follow-up)

• single-node/cross-chain/** (PR 3).
• single-node/fees/private_payments.parallel (640), single-node/block-building/block_building, single-node/proving/default_node, single-node/sequencer/runtime_config (PR 4).
• multi-node/slashing/** (PR 6): ~10 files with 1024 literals, incl. slashing/setup.ts and inactivity_setup.ts.
• multi-node/governance/setup.ts (640), multi-node/block-production/{setup.ts,blob_promotion}, multi-node/recovery/pipeline_prune, multi-node/invalid-attestations/** (PR 7).
• p2p/** (PR 8, incl. SHORTENED_BLOCK_TIME_CONFIG_NO_PRUNES's 640), composed/** (PR 9), automine/** (PRs 1-2).