Use non-pointer HistogramBucket slice

[PaurushGarg]

…esponse encoding

What this PR does:
This PR adds (gogoproto.nullable) = false to the repeated HistogramBucket buckets field - generating []HistogramBucket (values) instead of []*HistogramBucket (pointers). This is consistent with other repeated fields in the same proto (e.g., samples) that already use this annotation.

• Eliminates per-bucket heap allocation.
• Reduces per-bucket cost from ~40 bytes (32-byte struct + 8-byte pointer) to ~32 bytes (struct only)
• ~20% reduction in codec bucket memory for NH responses
• Reduces per-bucket heap overhead and cuts GC tracing by allocating contiguously.

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]
• docs/configuration/v1-guarantees.md updated if this PR introduces experimental flags

[yeya24]

Thanks. The change looks good. I ran some benchmark locally for this PR and the result looks good.

Buckets		ns/op	B/op	allocs/op
8	master	1,035	720	11
8	PR	834	640	2
8	Δ	−19%	−11%	−82%
32	master	3,177	1,696	35
32	PR	2,480	1,504	2
32	Δ	−22%	−11%	−94%
160	master	14,586	6,912	163
160	PR	11,157	5,728	2
160	Δ	−23%	−17%	−99%

[SungJin1212]

@PaurushGarg
lgtm, can you get the rebase?