[TraceQL Metrics] Add local disk caching for generator completed blocks

[mdisibio]

What this PR does:
Generators service metrics queries against recent data, which is done frequently and sensitive to user-facing performance. This adds a cache of query responses for local completed blocks which is typically the last 5-20 minutes.

The generators already have a similar cache for the metrics summary api, but this takes a different approach. Whereas the summary api cache is in-memory, this is disk-based and writes new caching files to the block folder. These files are unique to each request (query + params), and exist for the lifetime of the block. They aren't flushed to object storage, and they get automatically cleaned up when the block is deleted. (Note - this has precedent in the flushed file that is written to block folders to track their flushed status) Reasons to avoid the in-memory cache is that generators are already usually memory-intensive, and the cache is often inadequate as-is. Example block contents:

pwd = /tempo/generator/traces/single-tenant/blocks/single-tenant/
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/bloom-0
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/cache_query_range_14197044629227963207.buf
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/cache_query_range_18089096181261108941.buf
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/cache_query_range_433499706678913880.buf
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/cache_query_range_4989462974766826143.buf
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/data.parquet
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/flushed
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/index
  65beeff3-d9c5-49c1-b69e-8c7b022024f3/meta.json

There are additional changes in this PR which proved necessary:

• Generator RF1 block meta times weren't right. They always flushed traces with the timestamp of "now" so the metas were off by roughly trace_idle_period + trace_flush_period. This fixes it to flush proper times.
• As part of that, it also requires that the generator traces WAL has a real ingestion_slack time. This fixes the default to match ingesters of 2 minutes. NOTE - This is different than the slack time for metrics.

Which issue(s) this PR fixes:
Fixes #

Checklist

• Tests updated
• Documentation added
• CHANGELOG.md updated - the order of entries should be [CHANGE], [FEATURE], [ENHANCEMENT], [BUGFIX]