Add type header to WAL/Checkpoint records and use TSDB records

[codesome]

What this PR does:

Add a single-byte type header to WAL and Checkpoint records and way to handle old records concurrently. Additionally, replace existing proto records in WAL with Prometheus TSDB records.

Which issue(s) this PR fixes:
Fixes #2426

Checklist

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

[codesome]

PR is ready for review

[gouthamve]

I think we don't handle the migration well. I can't see where we are handling the errors returned by failed proto marshalling

[gouthamve]

Actually nvm, I missed the merr = nil bits :) It's all good! LGTM!

[codesome]

After some offline discussion with Goutham, we decided to add the Prometheus TSDB records in this PR itself to not have many versions or records, as the current record would be obsolete immediately after TSDB records are in.

[codesome]

I wrote a benchmark for WAL replay (will open PR after #2465 is merged as it depends on that) to compare the new records, and the results look good. Largely depends on Checkpoint/WAL ratio. The benchmark had equal number of samples in WAL and Checkpoint.

benchmark                    old ns/op      new ns/op      delta
BenchmarkWALReplay/#00-8     2027607295     1656442045     -18.31%

benchmark                    old allocs     new allocs     delta
BenchmarkWALReplay/#00-8     3050719        1142134        -62.56%

benchmark                    old bytes      new bytes      delta
BenchmarkWALReplay/#00-8     5943738312     5638734152     -5.13%

[pstibrany]

LGTM with some minor comments. processWAL method seems pretty convoluted, but is not part of this PR.

[pstibrany]

Tests are failing here now.