more accurate but slower

[WafiWadud]

just some small adjustments

[OmerZeyveli]

Hey, noticed a couple of issues in this PR:

• The to_fixed() scaling factor (10000000) seems excessively large and may cause overflow or loss of precision.

• Using 10000000 as the bit-width in check_overflow() effectively disables proper overflow detection.

These changes might actually reduce accuracy instead of improving it. Maybe worth revisiting the fixed-point scaling approach.