negative value cast to unsigned when calculating bytes

[thirtytwobits]

https://github.com/mattconte/tlsf/blob/master/tlsf.c#L1100 bytes - tlsf_size() where bytes < tlsf_size() yields a junk result in the tlsf_add_pool function.

Given a value of 40 bytes this leads to segfaults. Given a value of 4096 bytes this leads to https://github.com/mattconte/tlsf/blob/master/tlsf.c#L997 reporting that the buffer is too large. (this reproduced on aarch64).

I'd add a check that bytes is > tlsf_size() and document that the buffer provided must be greater than this value.