feat(embeddings): use stdlib array type for improved performance

[tonybaloney]

Implements #2059

Changes the default embedding request to use base64 responses (unless overriden).
Changes the default response parser to use the builtin array type to convert from the base64 to a float32 compact array then directly into a list.

4x faster than the current default (see issue with benchmark)
20% faster than with numpy, although I've left the numpy code in. The numpy code could be removed completely.

Benchmark	Min	Max	Mean	Min (+)	Max (+)	Mean (+)
Standard vs array	0.988	1.523	1.233	0.292 (3.4x)	0.313 (4.9x)	0.297 (4.1x)
Standard vs numpy	0.928	1.046	0.991	0.341 (2.7x)	0.370 (2.8x)	0.361 (2.7x)
numpy vs array	0.333	0.383	0.347	0.284 (1.2x)	0.316 (1.2x)	0.298 (1.2x)

[RobertCraigie]

Thanks for this!

I think you'll need to rebase against next to fix the test errors in this PR fyi

[tonybaloney]

I've rebased to next.

Regarding the defaults, it still has the same fallback to inspect the result. Adding the parameter is currently a side-effect of having numpy installed in the same venv, no?

[RobertCraigie]

Very nice, thank you!