Investigate Benchmarks

[prataprc]

BenchmarkVal2CollMap 50000 26696 ns/op 16416 B/op 2 allocs/op
BenchmarkVal2CollTyp 10000 174548 ns/op 49250 B/op 6 allocs/op

BenchmarkJson2CollStr 2000 635015 ns/op 1048647 B/op 5 allocs/op

BenchmarkCbor2CollMap 2000 787087 ns/op 1089654 B/op 8 allocs/op
BenchmarkCbor2CollTyp 500 2500523 ns/op 3269067 B/op 25 allocs/op

[prataprc]

strconv.AppendFloat, and, strconv.AppendInt avoid heap allocations, lowering
the memory pressure, hence suitable for tight loops. Except that, it does
appen() on the input byte-slice hence the slice itself escapes to heap. So
there is still 1 allocation that can't be avoided.

In case of CollateIint64 and CollateFloat64, there is an allocation for each call to AppendFloat/AppendInt.

Above benchmarks are on data that has integers and floats. So, we can't have a true ZERO allocation API to convert float to string and int to string.