Phase E: re-run the capacity benchmark with real shard-aware transport; finalize deferred approval thresholds

[brylie]

workspace-sharding.md §8/§10 explicitly defer these to real measurement
against Phase B's shard-aware transport (the #31 baseline only measured the
Phase-0 single-Y.Doc envelope).

• Re-run npm run benchmark:workspace:large against real per-shard
  transport: per-Document/per-Collection encoded state size, catalog
  size/refresh payload, SSE reconnect/resync behavior, cold load/sync
  latency, fan-out bytes per shard, server/client resource use,
  snapshot/restart time.
• Record a dated benchmark note per e2e-testing.md §6.
• Update workspace-sharding.md §10 with the Collection row-partition
  threshold, event-retention window, snapshot cadence, and state
  whether Document/Collection shard granularity holds at measured
  scale (reopening the design doc if it doesn't).

Done when: §10's deferred items are resolved with real measurement, not the
#31 projection, and the doc is updated accordingly.

[brylie]

Set Priority: unset → P1: only needs #120's real shard-aware transport to exist, not #121/#122 — the benchmark re-run can proceed once #120 lands, in parallel with #121/#122 rather than strictly after them.

[brylie]

Blocked by #120 (Phase B): the capacity benchmark needs real per-shard Y.Doc transport to measure against — it does not need #121 (SSE) or #122 (lazy-unload) to have landed first, so it can run in parallel with those once #120 is done.