Fix performance for large schema writes in V1Alpha1

[josephschorr]

This will hopefully reduce the flakiness of the E2E test suite, which writes large batches of definitions via the V1Alpha1 WriteSchema call

[vroldanbet]

LGTM. Left 2 nit suggestions, and a potential follow up to how we handle overlapping keys in CRDB

[vroldanbet]

nit: may confuse the reader, even though read should be nil

Suggested change

	return read, err
	return nil, err

[vroldanbet]

Suggested change

	return read, err
	return read, nil

[vroldanbet]

for very large namespaces (presumably the ones causing the flakiness we are attempting to fix in this PR) this would lead to many "touches" in the transactions table:

spicedb/internal/datastore/crdb/crdb.go

Lines 260 to 264 in ffdf521

	for k := range rwt.overlapKeySet {
	if _, err := tx.Exec(ctx, queryTouchTransaction, k); err != nil {
	return fmt.Errorf("error writing overlapping keys: %w", err)
	}
	}

@ecordell is there are reason we don't batch all those query operations instead of looping over the keyset?

[ecordell]

No real reason other than almost all requests should have only 1 overlap key

[vroldanbet]

Awesome, I think it would be relatively easy to turn that into a batch pgx request

[vroldanbet]

ugh sorry I didn't notice auto-merged was enabled 😭