Don't quantize resampled drawing coordinates coarser than necessary

1/8 is coarser than the actual 1/64 precision
of both VSFilter and libass (for \p1).

Moreover, quantization to exact 1/8 steps causes the output to have
up to 3 figures after the decimal point (e.g.: 0.125). If reducing
the output size were the goal, then rounding to 1/10 would trivially
produce shorter output *and* less quantization error, and even 1/100
would produce shorter output for uniformly distributed input values.

Current Lua automation scripts for Aegisub commonly round
drawing coordinates to 1/1000, so stick to that precision here.
This happens to be float_to_string's default, so just rely on that.

Author: astiob

src/resolution_resampler.cpp

@@ -75,7 +75,6 @@ namespace {
 					val = (val + shift_x) * scale_x;
 				else
 					val = (val + shift_y) * scale_y;
-				val = round(val * 8) / 8.0; // round to eighth-pixels
 				final += float_to_string(val);
 				final += ' ';
 				is_x = !is_x;