Implemented ldexpf in assembly (subnormal input/output returns zero)

[ZERICO2005]

ldexpf has been implemented in assembly:

• normal inputs are handled correctly, unless the output is subnormal
• subnormal inputs/outputs return zero and set ERANGE and FE_INEXACT
• zero/infinite/NaN inputs are handled correctly
• Underflow returns positive zero for better compatibility with Ti floats. Negative zero is only returned when the input is negative zero

subnormal inputs/outputs will be handled via eZ80sf routines for efficient normalization and correct rounding in the future.

Optimized frexpf, reduces size by 5 bytes:

• normal/subnormal: 1F faster
• inf/NaN: 2F slower
• zero: 7F + 1R + 1W + 1 slower (still faster than normal/subnormal)