perf: improve proto marshaler

[muktihari]

Context: MarshalBinary()

• Message Definition: Reduce bytes alloc by directly append to b. For this case, there is no benefit on using sync.Pool since the bytes size is known at front, using sync.Pool will only add overhead.
• Message: Use sync.Pool to reduce memory usage and alloc since the method might be called thousands times, there is not much different in sec/op but memory usage is way reduced

goos: darwin
goarch: amd64
pkg: github.com/muktihari/fit/proto
cpu: Intel(R) Core(TM) i5-5257U CPU @ 2.70GHz
                           │   old.txt    │               new.txt                │
                           │    sec/op    │    sec/op     vs base                │
MarshalMessageDefinition-4   731.2n ± 39%   242.2n ± 77%  -66.88% (p=0.000 n=10)
MarshalMessage-4             2.913µ ± 25%   2.724µ ± 23%        ~ (p=0.123 n=10)
geomean                      1.459µ         812.1n        -44.35%

                           │  old.txt   │               new.txt                │
                           │    B/op    │    B/op     vs base                  │
MarshalMessageDefinition-4   290.0 ± 0%   290.0 ± 0%        ~ (p=1.000 n=10) ¹
MarshalMessage-4             628.0 ± 0%   180.0 ± 0%  -71.34% (p=0.000 n=10)
geomean                      426.8        228.5       -46.46%
¹ all samples are equal

                           │  old.txt   │               new.txt               │
                           │ allocs/op  │ allocs/op   vs base                 │
MarshalMessageDefinition-4   2.000 ± 0%   2.000 ± 0%       ~ (p=1.000 n=10) ¹
MarshalMessage-4             81.00 ± 0%   78.00 ± 0%  -3.70% (p=0.000 n=10)
geomean                      12.73        12.49       -1.87%
¹ all samples are equal