Use flightCache for index cache implementation

[markusthoemmes]

Currently, there's a race between store, load and forget if indexes get looked up concurrently and change mid-way through the process. The r ace goes like this:

1. A: HEAD request → gets etag1, key = "url@etag1"
2. A: Once.Do runs: fetches, parses, calls store("url@etag1", idx), then sets urlToEtag[url] = etag1
3. B: HEAD request → gets etag2 (index was updated server-side), key = "url@etag2"
4. B: Once.Do runs: sees prev = etag1, calls forget("url@etag1") — which deletes indexes["url@etag1"]
5. A: calls load("url@etag1") → key not found → consistency error

Using the inFlight cache fixes that as values are not stored and loaded separately.

Separately, there's another race where we'd leak indexes with outdated etags if two HEAD requests would race outside of the critical section of adjusting the etag pointers. That's fixed through forgetting all "old" etag values for a given index URL as well.