NK_DTOA needs more documentation

Nobody has claimed this yet.

The current documentation states:

> You can define this to dtoa or your own double to string conversion implementation replacement. If not defined nuklear will use its own imprecise and possibly unsafe version (does not handle nan or infinity!).

The problem is, there is no such function. `dtoa` is neither in C standard library, C++ standard library or in POSIX. I just don't know what is the "contract" of this function.

Based on Nuklear's code:

https://github.com/Immediate-Mode-UI/Nuklear/blob/ca8aaf3f65e9fc8ac5d90fefc1204585add4b89b/nuklear.h#L7116-L7117

The function takes a buffer and a value of type `double` and returns the end of the string in the buffer.

*How large the buffer should be?* I checked the implementation and found out it is `NK_MAX_NUMBER_BUFFER`. This constant is documented but it is easy to lose it - NK_DTOA does not mention it and does not even specify how the function should look like (and there is no standard function to look into).

IMO it should be documented that:
- `NK_DTOA` relies on `NK_MAX_NUMBER_BUFFER`
- what the expected function prototype is
- what the expected behavior is (e.g. returns start or end of buffer)

Start in nuklear.h at the NK_DTOA implementation around lines 7116-7117, then compare it with the existing NK_MAX_NUMBER_BUFFER documentation. Document the expected function prototype, buffer-size relationship, and whether the return value points to the start or end of the generated string.

Written by the indexing model from the issue text.