improve the performance of trimming shape run cache

[dzhou121]

When shape run cache was trimmed with a large keep_ages, it's expensive to scan the whole cache.

This PR is trying to save the cache keys to a Vec of age registries, and when doing trim, pop the Vec according to keep_ages and those are the keys that needed to be removed from the cache.

[jackpot51]

Sorry for the delay in reviewing. I have one item I would like fixed.

[jackpot51]

Looks like Rust/build CI is failing

[UkoeHB]

You can potentially reuse a hashset popped earlier.

[dzhou121]

Can you please re-run the CI?

[jackpot51]

This looks good to me. Given that @UkoeHB also provided feedback, I'd like a re-review from them as well.

[UkoeHB]

Suggested change

	self.age_registries[index].remove(key);
	// update age
	*age = self.age;
	// register the key to the new age registry
	if let Some(keys) = self.age_registries.first_mut() {
	keys.insert(key.clone());
	}
	let prev_copy = self.age_registries[index].take(key);
	// update age
	*age = self.age;
	// register the key to the new age registry
	if let Some(keys) = self.age_registries.first_mut() {
	// Note: This is only valid so long as the PartialEq impl on ShapeRunKey checks value
	// equality.
	keys.insert(prev_copy.expect("age_registries should have entry if cache has entry"));
	}

[UkoeHB]

Suggested change

	// and remove the keys from cache saved in the registries
	while self.age_registries.len() as u64 > keep_ages {
	if let Some(keys) = self.age_registries.pop() {
	for key in keys {
	self.cache.remove(&key);
	}
	}
	}
	// and remove the keys from cache saved in the registries
	let mut recovered_keys: Option<HashSet<ShapeRunKey>> = None;
	while self.age_registries.len() as u64 > keep_ages {
	if let Some(keys) = self.age_registries.pop() {
	for key in keys.drain() {
	self.cache.remove(&key);
	}
	if recovered_keys.is_none() {
	recovered_keys = Some(keys);
	}
	}
	}

[UkoeHB]

Suggested change

	self.age_registries.insert(0, HashSet::default());
	self.age_registries.insert(0, recovered_keys.unwrap_or_default());