RuleGroup: use stdlib for json decoding, optimize decoding

[liggitt]

cc @bwplotka

part of #2105

1. Adds a benchmark for RuleGroup decoding
2. Switches to stdlib
3. Avoids decoding rule bytes multiple times
4. Avoids struct copies

Allocs and bytes consumed are significantly better than json-iter on both Go 1.26 and 1.27

CPU is worse on Go 1.26, better on Go 1.27.

goos: darwin
goarch: arm64
pkg: github.com/prometheus/client_golang/api/prometheus/v1
cpu: Apple M4 Pro
                       │    main     │              go1.26.6               │              go1.27.1              │
                       │   sec/op    │    sec/op     vs base               │   sec/op     vs base               │
RuleGroup/streaming-14   944.5µ ± 1%   1551.7µ ± 0%  +64.28% (p=0.002 n=6)   787.3µ ± 3%  -16.64% (p=0.002 n=6)
RuleGroup/unmarshal-14   885.9µ ± 1%   1521.7µ ± 1%  +71.76% (p=0.002 n=6)   722.0µ ± 2%  -18.50% (p=0.002 n=6)
geomean                  914.7µ         1.537m       +67.98%                 754.0µ       -17.58%

                       │     main      │              go1.26.6               │              go1.27.1               │
                       │     B/op      │     B/op      vs base               │     B/op      vs base               │
RuleGroup/streaming-14   1293.9Ki ± 0%   832.7Ki ± 0%  -35.64% (p=0.002 n=6)   620.5Ki ± 0%  -52.04% (p=0.002 n=6)
RuleGroup/unmarshal-14    998.0Ki ± 0%   578.3Ki ± 0%  -42.05% (p=0.002 n=6)   363.5Ki ± 0%  -63.57% (p=0.002 n=6)
geomean                   1.110Mi        693.9Ki       -38.93%                 475.0Ki       -58.20%

                       │     main     │              go1.26.6              │              go1.27.1              │
                       │  allocs/op   │  allocs/op   vs base               │  allocs/op   vs base               │
RuleGroup/streaming-14   27.396k ± 0%   9.614k ± 0%  -64.91% (p=0.002 n=6)   3.868k ± 0%  -85.88% (p=0.002 n=6)
RuleGroup/unmarshal-14   27.314k ± 0%   9.601k ± 0%  -64.85% (p=0.002 n=6)   3.850k ± 0%  -85.90% (p=0.002 n=6)
geomean                   27.35k        9.607k       -64.88%                 3.859k       -85.89%

[bwplotka]

Thanks!