API: Add indexStatsNames to create field names for content stats

[rdblue]

This introduces a new TypeUtil method to index fields in a type that produces names that can be used for stats fields in content stats. These names prefer short names, like locations.lat instead of locations.element.lat and uses underscores instead of . to join names into a single string.

This is exposed through TypeUtil to avoid making more of IndexByName public.

[wombatu-kun]

Heads up: the red core-tests on this PR is a real regression from this change, not a flaky test. Root cause and a suggested fix inline on byId().

[wombatu-kun]

Switching byId() from build() to buildKeepingLast() also changes the two existing callers of byId(), TypeUtil.indexNameById and indexQuotedNameById. Schema's constructor relies on indexNameById to reject duplicate field IDs: it calls lazyIdToName() -> indexNameById() -> byId(), and the ImmutableMap keyed by field ID used to throw on a duplicate key. With buildKeepingLast() a schema with two fields sharing an ID is now silently accepted, which is why TestNameMapping.testFailsDuplicateId fails on both JDK 17 and 21 ("Expecting code to raise a throwable").

The short-name-overrides-full-name behavior is only needed for the new indexStatsNames path, so gating buildKeepingLast() behind useShortNames keeps the existing validation intact:

Suggested change

	if (useShortNames) {
	shortNameToId.forEach((key, value) -> builder.put(value, key));
	}
	return builder.buildKeepingLast();
	if (useShortNames) {
	// short names intentionally override the full name for the same field ID
	shortNameToId.forEach((key, value) -> builder.put(value, key));
	return builder.buildKeepingLast();
	}
	return builder.build();