Looks like mp_hal_pin_low() and mp_hal_pin_high() in rp2/mphalport.h have to be fixed to use gpio_set_mask64() and gpio_clr_mask64() instead of the 32bit versions. Someone with a RP2350B could implement and test it. It may have an impact on Neopixel timing.

Turns out the reason the 64-bit versions didn't work is because you need to explicitly typecast to a uint64_t. Will update my PR, since I think the mask functions are faster than gpio_put() (though I didn't see any timing issues).

though I didn't see any timing issues

I will compare it to the previous state tomorrow. I still have the old test data from rp2040. The difference should be small, like a few clock cycles per output phase.
Edit: Looking into the code, lacking an RP2350B the test make no sense.

Actually, would it even matter if it takes longer? The bitstream implementation (which the NeoPixel driver uses) just waits until it's time to toggle the pin state:

So as long as setting the pin high/low doesn't take longer than the time between pulses (shortest is 400ns), I don't think it matters if this change results in a couple extra clock cycles.

Probably it will not matter. Especially not for neopixel, which has quite relaxed timing requirements.

Testing gives sometimes unexpected results. I tested the bitstream timing with a unmodified RP2350A and the differ from the rp2040. So it needs a small adjustment. I made a change to the compensation value as of:

#if PICO_RP2350
#define MP_HAL_BITSTREAM_NS_OVERHEAD  (5)
#else
#define MP_HAL_BITSTREAM_NS_OVERHEAD  (9)
#endif

Maybe you could include this change into your PR, making it a single commit for about the same topic.

All timings made with a bit tuple of (1000, 1000, 1000, 1000) and \x55\x55 as pattern.
Timing with 9ns compensation at 125MHz clock:

Timing with 5ns compensation at 125MHz clock:

Timing with 5 ns compensation at 150MHz clock (default for RP2350):

For comparison a figure for RP2040:
Timing with 9ms compensation at 125MHz:

For comparison the timing for RP2350, RISCV. The 150Mhz figures are good, the ones for 125MHz and 250Mhz are worse. They would need a larger compensation value. The value for 250 Mhz is surprising. Usually the timing got better at higher frequencies.

150MHz, 5ns, compensation:

125MHz, 5 ns compensation:

PR updated with your suggested timings!