Tempo: Refactor Trace By ID and Metrics Summary

[joe-elliott]

What this PR does:
Refactors 2 more endpoints to use the new frontend pipeline items. The final one will be the new query range endpoint. Trace by ID was more difficult than expected. It has a few unique features that make it harder to work with:

• Passes proto instead of json
• Changes the encoded object. All other paths propagate the same object from querier to outside
• Has a special middleware that occurs after recombination (deduper)
• Does not consider 404s an error

Testing is looking good for no regressions.

Other changes:

• Removes trace by id hedging. We have found it to be an antipattern.
• Fixes 429s being returned as 499s and adds an integration test
• Removes the addition of a tenant in multitenant trace id lookup. We are not using it and it's difficult to fit into the new pipeline. Ignoring for now.
• Reduce memory consumption in the frontend on trace by id lookup by pushing the deduper and content-type marshalling logic into the combiner. This prevents an unnecessary 2x marshalling/unmarshalling of a trace.

Which issue(s) this PR fixes:
Works toward #3413

Checklist

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

[knylander-grafana]

Doc updates look good. Thank you for updating the docs.

[yvrhdn]

LGTM overall 👍 Left some minor comments

[yvrhdn]

I'm fine with merging this 👍 just got some minor questions/issues, nothing blocking