refactor(storage): carry the cache hit through open instead of unwrapping it

[BryanFRD]

Closes #566

Store::open for a bucket now returns the cache alongside the reader when the reader came from it ((reader, hit)), and the raw-object branch reopens through hit. That removes the from_cache flag and the cache.as_ref().unwrap() behind it, and the two identical Object::Remote arms collapse into one. The invariant is now in the types.

No behaviour change. I added two tests that run against the in-process bucket stub, so they need no emulator:

• a raw object is read from the bucket cold (Object::Remote), the cache is filled behind the read, and the next read is Object::Raw with the same bytes;
• with no cache configured, a raw object is always Object::Remote.

Both also pass against the code on main, which is the point: they pin the behaviour the refactor must keep. The only existing test of this branch, a_cached_download_is_the_same_download, needs a bucket emulator and only runs in CI.

[ferrfleet]

The refactor doesn't change behaviour: hit is Some only when the cache served the reader, and that's the condition from_cache used to check, so the unwrap goes away with nothing else changing. The new warm-cache test is sound. Cache::fill writes the .b3 digest before it renames the object into place, so polling for the object file can't race the integrity check in open.

One formatting nit inline.

[ferrfleet]

Nit: this fits within the default 100-column width, so cargo fmt --check will want it on one line.

Suggested change

	assert!(
	copied.exists(),
	"the copy is filled in behind the first read"
	);
	assert!(copied.exists(), "the copy is filled in behind the first read");