[FEA] Support reduced precision in IVF-PQ index building

[cjnolet]

From @tfeher:

Currently everything is upconverted to fp32. Keep input precision. Use optimized k-means for the same precision / consider downconverting while subsampling.

This addresses part of #1675

[tfeher]

There are two separate tasks here, which can be solved independently:

• Enable handling fp16 input dataset for IVF-PQ.
• Do not upconvert fp16 when subsampling the index (see the issue description above). This depends on [FEA] Support reduced precision in balanced k-means  #1892