Avoid hacky workaround for with_capacity

[jonhoo]

When we have to allocate a new table, we want to continue with factor-of-two resizing when possible. Unfortunately, hashbrown::RawTable::with_capacity (I think) allocates more or less the capacity we give it, rather than rounding up. So, we resort to this workaround:

griddle/src/lib.rs

Lines 746 to 749 in c68eb4c

	// We want to favor factor-of-two growing, if that's sufficient, so we don't
	// use with_capacity (which allocates _exactly_ what we give).
	let mut new_table = RawTable::new();
	new_table.reserve(need_cap, |_| unreachable!());

It'd be nice to get rid of that. Benchmarks may tell us that with_capacity works just fine..

[Amanieu]

reserve and with_capacity are equivalent: they both take the capacity (number of elements you can insert before resizing) as an argument, not the number of buckets.

[Amanieu]

They also both round up to the next power of two.

[jonhoo]

Hmm, I could be wrong, but that doesn't quite seem like the case from the code? While reserve does eventually call try_with_capacity, it computes the additional capacity to request using a usize::max here, which seems to do some rounding up. Is that logic not relevant/important?

[Amanieu]

new_items may be less than full_capacity at this point if the rehash path was reached because all of our free space is used up by tombstones. The usize::max here is to force us to use the next size in that case.