NSF driver: optimize tempo calculation

[Gumball2415]

Whenever the driver calculates the next tempo tick interval through ft_restore_speed, it has to do this for unfixed-tick BPM:

player.s L1738

	lda var_Tempo_Accum
	adc var_Tempo_Dec
	sta var_Tempo_Accum
	lda var_Tempo_Accum + 1
	adc var_Tempo_Dec + 1
	sta var_Tempo_Accum + 1
	sec
	lda var_Tempo_Accum
	sbc var_Tempo_Modulus
	sta var_Tempo_Accum
	lda var_Tempo_Accum + 1
	sbc var_Tempo_Modulus + 1
	sta var_Tempo_Accum + 1

in pseudocode, this would mean Accumulator += RegionSpeedDivider; Accumulator -= Modulus;.

At this point, the modulus and the decrement would be constant after calling ft_calculate_speed. it would be optimal to cache the difference of RegionSpeedDivider and Modulus (DecrementMod = RegionSpeedDivider - Modulus;), and then add it to the Accumulator as Accumulator += DecrementMod;.

[Gumball2415]

correction: var_Tempo_Dec actually holds the region speed divider.