math/big: make Float implement encoding.TextMarshaler and encoding.TextUnmarshaler interfaces

[kostya-sh]

Both big.Int and big.Rat have MarshalText and UnmarshalText methods to implement two interfaces from encoding package: TextMarshaler and TextUnmarshaler.

big.Float should implement these interfaces as well

[gopherbot]

CL https://golang.org/cl/13882 mentions this issue.

[kostya-sh]

Implementing these interfaces is less trivial than it seems.

Fast and looseless implementation could use hexadecimal notation with 'p' (binary) exponent. But it depends about the expectations of MarshalText: Should it always return a string representation that is a "standard" floating-point number that might be read by another system? In that case, the 'p' notation might not be option.

An implementation that is more similar to float32/64 can use 'g' with precision -1 (once implemented) but it's a very expensive conversion. Additionally currently there is no way to parse a string without loosing precision. Parse's semantic may actually not be correct as is: It perhaps should parse w/o rounding error if the precision is 0.

[griesemer]

Decided to keep it simple for now and implement this simply via the 'g' with precision -1 format. Also, no recording of precision or accuracy. This makes this work for json w/o trouble.

[gopherbot]

CL https://golang.org/cl/15040 mentions this issue.