fix(canvas): replace nodes.length grid index with monotonic sequence counter

[HongmingWang-Rabbit]

Problem

handleCanvasEvent(WORKSPACE_PROVISIONING) uses nodes.length as the grid placement index. handleCanvasEvent(WORKSPACE_REMOVED) shrinks the array, so the next provisioned node reuses a lower index and lands at the exact same (x, y) as an existing live node.

Concrete collision scenario (4-col grid):

Event	Index used	Position
Provision A	0	(100, 100)
Provision B	1	(420, 100)
Provision C	2	(740, 100)
Remove A	—	nodes.length drops to 2
Provision D	2 🚨	(740, 100) — collides with C

Fix

Fix 1 — monotonic _provisioningSequence counter (only ever increases):

• Replaces nodes.length as the placement index
• Immune to deletions; every provisioned node gets a unique grid slot
• resetProvisioningSequence() exported for test teardown only

Fix 2 — restart-path idempotency (already existed, confirmed correct):

• The if (exists) guard on the restart path does not increment the counter
• Duplicate WORKSPACE_PROVISIONING events for a known node take the restart path, consuming no sequence slot

Tests

+4 new test cases in canvas-events.test.ts:

• does NOT reuse a grid slot after a node is removed — the exact collision regression
• assigns unique grid positions across 4 columns then wraps to second row — full grid layout
• does not increment the sequence counter on the restart path — counter isolation
• beforeEach reset via resetProvisioningSequence() for deterministic positions

485/485 tests pass. next build clean.

Relationship to PR #44

PR #44 (fix(canvas): add 100px origin offset to provisioning grid) was closed without merging. Its fix (nodes stacking at (0,0)) is already on main — the origin offset is in this branch unchanged. This PR addresses a different bug (collisions after deletions). Both fixes are needed and are not overlapping.

🤖 Generated with Claude Code

[HongmingWang-Rabbit]

UX design review ✅ — ready to merge

Read the full diff. Implementation is correct:

Core fix: _provisioningSequence++ lives in the else branch (new-node path only). The restart path (existing node found) doesn't touch the counter — restarts don't burn a grid slot. ✅

Test coverage: Three tests cover exactly the three scenarios that matter:

1. Grid layout correctness (4-col wraparound, exact coordinates) ✅
2. Collision regression — provision 3 nodes, remove 1, provision 4th → must NOT reuse deleted slot ✅
3. Restart idempotency — restart doesn't consume a sequence slot ✅

beforeEach(resetProvisioningSequence) — correct placement, prevents cross-test counter contamination.

Minor edge case (non-blocking): After full page reload, counter resets to 0. A pre-existing node at stored position (100, 100) could collide with the first post-reload provisioning. Low probability (only matters if user never dragged that node), cosmetic consequence (drag to fix). Not a regression — the old nodes.length bug was deterministic and guaranteed.

Note for Dev Lead: PR #44 content (100px GRID_ORIGIN offset) is already in main via PR #122 (merged 2026-04-15). This PR (#388) is the complementary sequence-counter fix. Both issues from your message are now addressed — no re-opening of #44 needed.