currently this function causes a crash in debug mode when used with pins greater than
32 (available on the rp235xb variant) because it overflows when doing
the shift. this commit applies the offset to pin value before doing the
shift. this assumes that when using a pin greater than 32 the GPIOBASE
has already been set to 16.

embassy_rp: improve calculate_pio_clock_divider

defmt::assert gives a compile error in const context.

Ensure that the fractional part of the clock divider is accurately
calculated. This does additional u32 division/mod operations, which I
think is better than using u64 for its extra precision.

Also:
- Add additional asserts to catch out-of-bounds results.
- Make the version that accepts the system clock as a parameter
const and public, so if anyone wants to hardcode the system
frequency they can avoid any runtime computation.
- Remove the `inline` attributes because the function has grown quite a
bit.

The test will fail, because the current implementation doesn't calculate
the fractional part.