Optimization: avoid cast to double in src_short_to_float

[jgcodes2020]

The current implementation of src_short_to_float looks like this:

libsamplerate/src/samplerate.c

Lines 427 to 435 in 2ccde95

	void
	src_short_to_float_array (const short *in, float *out, int len)
	{
	for (int i = 0 ; i < len ; i++)
	{ out [i] = (float) (in [i] / (1.0 * 0x8000)) ;
	} ;
	return ;
	} /* src_short_to_float_array */

(1.0 * 0x8000) yields a double and so the division is done in doubles, requiring a few extra casts. This can be avoided by doing it in floats like this:

void
src_short_to_float_array (const short *in, float *out, int len)
{
	for (int i = 0 ; i < len ; i++)
	{	out [i] = (float) (in [i] / (1.0f * 0x8000)) ;
		} ;


	return ;
} /* src_short_to_float_array */

This should yield identical results because 1/32768 is an exact power of 2.