Add batch_observe methods to Histogram and Summary

[MahmoudAshraf97]

This PR adds batch methods for observing a list of values that become available at once.

Motivation

The use case for this is in high-throughput ML inference systems where some metrics are calculated for the whole batch at once such as queue time spent per element, we usually looped over the elements and used Histogram.observe, but each call involved python loop overhead and acquired two locks per call

Speedup

Python 3.12.3, x86_64 (AMD EPYC 9334), default buckets, best of 5. Loop = n calls to observe().

Single process (MutexValue)

n	Summary loop	Summary batch	speedup	Histogram loop	Histogram batch	speedup
16	6.7 us	0.7 us	9.9x	11.1 us	3.5 us	3.2x
64	26.0 us	1.0 us	25.3x	45.4 us	7.6 us	6.0x
256	105.1 us	2.6 us	41.1x	182.6 us	23.0 us	8.0x
1024	416.8 us	8.5 us	49.1x	729.8 us	87.9 us	8.3x

Multiprocess mode (MultiProcessValue, mmap write per inc)

n	Summary loop	Summary batch	speedup	Histogram loop	Histogram batch	speedup
16	23.9 us	1.8 us	13.4x	29.6 us	8.1 us	3.7x
256	382.0 us	3.8 us	100.4x	477.6 us	28.7 us	16.6x

It might seem that despite the large speedup, it is still in order of microseconds, but in our use case, these microseconds block the asyncio event loop for hundreds of other tasks, and these batched methods increased the total system throughput by 2% just by improving observability tools, which is meaningful given the current GPU prices

I'm also open to extend the regular observe methods to accept lists or sequences, but this is left for the maintainers to decide

TIA