fix: reduce allocations in the Time Window Block Selector

[javiermolinar]

What this PR does:

I noticed a huge amount of objects allocated during compactions from the Sprintf method

This PR reuses a string buffer to reduce these allocations

❯ benchstat old.txt new.txt
goos: darwin
goarch: arm64
pkg: github.com/grafana/tempo/tempodb
cpu: Apple M3 Pro
              │   old.txt   │               new.txt               │
              │   sec/op    │   sec/op     vs base                │
BlockSelector   60.07m ± 1%   25.37m ± 2%  -57.77% (p=0.000 n=10)

              │   old.txt    │               new.txt                │
              │     B/op     │     B/op      vs base                │
BlockSelector   43.74Mi ± 0%   28.23Mi ± 0%  -35.45% (p=0.000 n=10)

              │   old.txt   │               new.txt               │
              │  allocs/op  │  allocs/op   vs base                │
BlockSelector   800.0k ± 0%   700.0k ± 0%  -12.50% (p=0.000 n=10)

Checklist

• Tests updated

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

[joe-elliott]

this is a very tiny %age of all compactor allocations and the previous implementation had a lot of clarity on what the group/order/hash strings looked like. the current implementation loses this a bit

i'm fine with moving forward with this, but we need the new implementation to read as cleanly as the old given that the value gained is so small