This commit eliminates all allocations during marshalling (except for
the final []byte slice allocation, which is unavoidable). This matches
the behavior of encoding/json. It also prevents the vwPool from leaking
writers and includes a few small optimizations to the value_writer.

Other Changes:
   * bsoncodec: reduce use of reflect.Value.Interface()
   * bson: use a buffer pool instead of SliceWriter to eliminate 2 allocs
   * bsonrw: use references to slice indexes

This work builds off of mongodb#1313
and uses the BenchmarkCode* suite added by that PR.

The combined performance improvement is below:

```
goos: darwin
goarch: arm64
pkg: go.mongodb.org/mongo-driver/bson
                      │ base.20.txt  │             new.20.txt              │
                      │    sec/op    │   sec/op     vs base                │
CodeUnmarshal/BSON-10    3.177m ± 1%   2.202m ± 1%  -30.68% (p=0.000 n=20)
CodeMarshal/BSON-10     2922.7µ ± 0%   757.1µ ± 2%  -74.10% (p=0.000 n=20)
geomean                  3.047m        1.291m       -57.63%

                      │ base.20.txt  │               new.20.txt               │
                      │     B/s      │      B/s       vs base                 │
CodeUnmarshal/BSON-10   582.5Mi ± 1%    840.4Mi ± 1%   +44.26% (p=0.000 n=20)
CodeMarshal/BSON-10     633.2Mi ± 0%   2444.3Mi ± 2%  +286.03% (p=0.000 n=20)
geomean                 607.3Mi         1.400Gi       +135.99%

                      │ base.20.txt  │              new.20.txt              │
                      │     B/op     │     B/op      vs base                │
CodeUnmarshal/BSON-10   4.219Mi ± 0%   4.148Mi ± 0%   -1.69% (p=0.000 n=20)
CodeMarshal/BSON-10     2.818Mi ± 3%   1.630Mi ± 0%  -42.16% (p=0.000 n=20)
geomean                 3.448Mi        2.600Mi       -24.59%

                      │  base.20.txt   │              new.20.txt              │
                      │   allocs/op    │  allocs/op   vs base                 │
CodeUnmarshal/BSON-10      230.4k ± 0%   220.7k ± 0%    -4.21% (p=0.000 n=20)
CodeMarshal/BSON-10     94066.000 ± 0%    1.000 ± 0%  -100.00% (p=0.000 n=20)
geomean                    147.2k         469.7        -99.68%
```