[BUG] Inconsistent use of IVF-PQ specialized types

[cjnolet]

I noticed the IVF-PQ specializations don't seem to be getting used consistently throughout the codebase, which I think is contributing to the growing build time.

Here's the (value_t, index_t) type pairs that are being used in each source file:

Search:

• Specializations: (float, int64_t), (float, uint32_t), (float, uint64_t)
• Googletests: (float, int64_t)
• Bench: (float, int64_t), (float, uint32_t), (int8_t, int64_t), (uint8_t, uint32_t)
• Runtime: (float, uint64_t), (int8_t, uint64_t), (uint8_t, uint64_t)

Any types being used which are not being specialized are going to cause that call to be recompiled each time it is used. We should make sure we're at least specializing the union of these types if not trying to see where we can prune some out.

[cjnolet]

Linking #1201