fix: stabilize Space drag projections

[DeadWaveWave]

💡 Change Scope

• Small Change: Fast feedback, localized UI/logic, low-risk.
• Large Change: New feature, cross-boundary logic, runtime-risk (persistence, IPC, lifecycle, recovery).

📝 What Does This PR Do?

Fixes two regressions in Workspace Canvas node/Space interaction:

• keeps the active window stable while slowly crossing or following a Space edge;
• previews the required Space expansion while a window is still held over a crowded Space, instead of revealing the expansion only after release.

The drag pipeline now separates per-frame primary constraints from throttled secondary peer/Space layout. Live Space geometry is a reversible UI projection; release still recomputes the authoritative ownership and geometry from durable state.

---

🏗️ Large Change Spec (Required if "Large Change" is checked)

1. Context & Business Logic

The previous drag throttling skipped the whole projection during slow edge movement, causing raw and constrained positions to alternate. The follow-up live path then intentionally returned unchanged Space geometry, so crowded-Space expansion was invisible until mouseup.

This change runs primary active-window constraints on every coalesced frame, keeps peer/Space reflow in the existing secondary lane, and derives expanded preview frames without persisting them. Capacity lower-bound and postcondition checks avoid the expensive bounded solver for known-overflow layouts.

2. State Ownership & Invariants

• Durable Space rects and membership remain owned by spacesRef / persistence and are written only by the release/finalization path.
• Ordinary node-drag Space previews are renderer-derived and reversible; they cannot be committed as Space-dominant drag state.
• The active dragged window always uses the primary projection. Secondary peer layout and auto-resize cannot overwrite it.
• A target Space change bypasses secondary distance/time throttling, preventing stale peer or Space previews.

3. Verification Plan & Regression Layer

• Unit: live/primary/final projection invariants, target switching, preview ownership isolation, Child Space/ancestor expansion, packed 81-window performance budget.
• Hook integration: secondary preview survives primary-only frames and cannot trigger Space persistence.
• Electron E2E: mouse-held crowded-Space expansion with durable state unchanged before release; slow Space-edge drag stability; Child Space and push-away regressions.
• Full gate: pnpm pre-commit exited 0 (237 E2E passed, 47 conditional skips, 3 unrelated flaky cases passed on retry).

---

✅ Delivery & Compliance Checklist

• My code passes the ultimate gatekeeper: pnpm pre-commit is completely green.
• I have signed the CLA if required (see CLA.md).
• I have included new tests to lock down the behavior (or explicitly stated why it's untestable).
• I have strictly adhered to the DEVELOPMENT.md architectural boundaries.
• I have attached a screenshot or screen recording (if this touches the UI).
• I have updated the documentation accordingly (if adding a feature or changing a contract).

📸 Screenshots / Visual Evidence

The interaction is covered by Electron E2E while the mouse remains held: the DOM Space frame expands before release, persistence remains at its baseline, and release commits the final non-overlapping layout. No review-only media was committed to the repository.