Reuse batchesBuf across series via BufferCarrier

[PaurushGarg]

What this PR does:
Each mergeIterator allocates its own batchesBuf (~1200 bytes) at creation time. For engines that allocate all iterators upfront (without series batching), this adds to the heap - while only one buffer is actively used at a time since series are processed sequentially within a shard.

This PR adds a bufProvider interface that allows an upstream caller to pass a shared buffer through s.Iterator(carrier). The first series allocates the buffer and writes it back to the carrier; all subsequent series reuse it. When no carrier is passed (nil), each iterator allocates its own buffer as before.

The companion promql-engine PR passes a BufferCarrier from the matrixSelector, enabling the reuse.

Benchmark
Added BenchmarkNewChunkMergeIterator_BufferCarrier which compares:

• no_carrier: creates 10K iterators passing nil (current behavior — each allocates own buffer)
• with_carrier: creates 10K iterators passing a shared carrier (new behavior — one buffer reused)

Results here shows -13.1% memory, -10K allocations, no speed regression. Savings are ~1280 bytes/series.

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

[PaurushGarg]

 
dev-dsk-paurushg-2a-87a5ee42 %   go test ./pkg/querier/batch/ -bench=BenchmarkNewChunkMergeIterator_BufferCarrier -run=^$ -benchmem -count=5

goos: linux
goarch: amd64
pkg: github.com/cortexproject/cortex/pkg/querier/batch
cpu: AMD EPYC 7R13 Processor
BenchmarkNewChunkMergeIterator_BufferCarrier/no_carrier-32         	       1	1238926825 ns/op	97765928 B/op	  820017 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/no_carrier-32         	       1	1241601111 ns/op	97760288 B/op	  820006 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/no_carrier-32         	       1	1231510838 ns/op	97760448 B/op	  820004 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/no_carrier-32         	       1	1228693033 ns/op	97765912 B/op	  820010 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/no_carrier-32         	       1	1235855774 ns/op	97760560 B/op	  820005 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/with_carrier-32       	       1	1237800866 ns/op	84961656 B/op	  810007 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/with_carrier-32       	       1	1235080586 ns/op	84961672 B/op	  810006 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/with_carrier-32       	       1	1233754924 ns/op	84961560 B/op	  810005 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/with_carrier-32       	       1	1230165453 ns/op	84961448 B/op	  810004 allocs/op
BenchmarkNewChunkMergeIterator_BufferCarrier/with_carrier-32       	       1	1223676402 ns/op	84961448 B/op	  810004 allocs/op
PASS
ok  	github.com/cortexproject/cortex/pkg/querier/batch	12.359s

(26-08-19 22:42:08) <0> [~/workplace/bench/cortex]