feat(ui): batch relationship value fetches to reduce N*M API requests

[jhb-dev]

What?

Introduces a RelationshipValueCacheProvider that batches document title fetches across all RelationshipInput field instances. Instead of each relationship field making its own REST request on mount, all fields that mount in the same render tick have their ID lookups collected and fired as a single request per collection.

Why?

When relationship fields are placed inside arrays (or blocks), each row's relationship field independently fetches its referenced document title via REST. With N array rows and M relationship fields per row, this produces N*M parallel requests. This causes:

• Severe UI performance degradation on documents with many array rows
• MongoDB connection exhaustion on Vercel (hitting the 500 connection limit)
• Unnecessary request bursts when drag-reordering array items

How?

• New RelationshipValueCacheProvider (packages/ui/src/providers/RelationshipValueCache/index.tsx) uses queueMicrotask to collect all getDoc() calls within a single tick, then fires one batched request per collection using where: { id: { in: [...allIds] } }
• Results are cached by collection:locale:id key so subsequent renders don't refetch
• Cache is invalidated on document delete events, updated on document save events, and cleared entirely on locale changes
• Collection fetches within a batch run in parallel via Promise.all
• RelationshipInput checks the shared cache before requesting, and delegates uncached IDs to the batching provider
• Provider is wired into the Root provider tree inside DocumentEventsProvider

Fixes #13329

[jhb-dev]

The failing CI check ("E2E - versions 1/3") is a pre-existing flaky test on main — unrelated to the changes in this PR. The test (test/versions/e2e.spec.ts:548 — autosave "Create new" navigation timeout) also fails on the latest main commit.

I'd appreciate a review from the team to get your thoughts on this approach before moving forward.