Optimize JSON parsing

[S7evinK]

Based on neilalexander/harmony@586b083 with minor changes.

Benchstat vs main:

goos: linux
goarch: amd64
pkg: github.com/matrix-org/gomatrixserverlib
cpu: Intel(R) Core(TM) i7-7700HQ CPU @ 2.80GHz
                              │   old.txt    │               new.txt                │
                              │    sec/op    │    sec/op     vs base                │
ParseLargerEvent-8               56.98µ ± 5%   49.62µ ±  3%  -12.91% (p=0.000 n=10)
ParseSmallerEvent-8              52.12µ ± 5%   47.13µ ± 14%   -9.58% (p=0.019 n=10)
ParseSmallerEventFailedHash-8   109.01µ ± 6%   98.07µ ±  4%  -10.04% (p=0.000 n=10)
ParseSmallerEventRedacted-8      83.67µ ± 3%   73.17µ ±  3%  -12.56% (p=0.000 n=10)
geomean                          72.14µ        64.00µ        -11.28%

                              │   old.txt    │               new.txt                │
                              │     B/op     │     B/op      vs base                │
ParseLargerEvent-8              29.17Ki ± 0%   16.75Ki ± 0%  -42.58% (p=0.000 n=10)
ParseSmallerEvent-8             23.25Ki ± 0%   16.16Ki ± 0%  -30.51% (p=0.000 n=10)
ParseSmallerEventFailedHash-8   40.19Ki ± 0%   26.05Ki ± 0%  -35.18% (p=0.000 n=10)
ParseSmallerEventRedacted-8     33.75Ki ± 0%   19.69Ki ± 0%  -41.65% (p=0.000 n=10)
geomean                         30.97Ki        19.30Ki       -37.67%

                              │  old.txt   │              new.txt               │
                              │ allocs/op  │ allocs/op   vs base                │
ParseLargerEvent-8              209.0 ± 0%   126.0 ± 0%  -39.71% (p=0.000 n=10)
ParseSmallerEvent-8             192.0 ± 0%   132.0 ± 0%  -31.25% (p=0.000 n=10)
ParseSmallerEventFailedHash-8   380.0 ± 0%   263.0 ± 0%  -30.79% (p=0.000 n=10)
ParseSmallerEventRedacted-8     282.0 ± 0%   168.0 ± 0%  -40.43% (p=0.000 n=10)
geomean                         256.1        164.6       -35.70%

[S7evinK]

Going to ignore the linter errors for now, those should be fixed in a different PR.