Previous work added the GPT2 static instance. However, the new
instance was still referring to the GPT1 clock gate. This commit
corrects that bug. We expose the GPT2 clock methods from the CCM,
add a new clock gate enumeration, then correctly instantiate the
GPT instances with the independent clock gates.

The commit also correct for an unused lifetime in the gpt typdefs.

Working through the new peripheral initialization API, I realized
that the GPIO driver wasn't amenable for the new API. The driver
was using static, mutable memory internally, which would make it
difficult to adapt to the new board-based peripheral initialization.
This commit tries to refactor the GPIO driver. I took the opportunity
to simplify some of the code, and address some potential bugs in the
implementation. I'll call out the possible issues in code review.

I tried to follow the "don't model the STM32 GPIO driver" guidance I
found in the code. This refactored driver should be a good compromise
that lets us move towards the board-based peripheral init API.

Teensy 4 light continues to blink from this commit. Kernel size
before this commit:

  make -C boards/teensy40

   text	   data	    bss	    dec	    hex
  62257	   9356	  12148	  83761	  14731

Size after:

   text	   data	    bss	    dec	    hex
  60853	   7884	  13620	  82357	  141b5

The commit lays the foundation for the board-based peripheral
initialization API. TODOs from this commit:

- remove static CCM instance
- remove static IOMUXC_SNVS instance
- fix UART clock control

We need to disable _all_ UART clocks at the clock gates before we
change the UART clock root. Before this commit, the configure method
would disable the clock gate for the current instance, configure the
clocks for _all_ UART instances, then re-enable its own clock. Though
the setting was the same for each configure call, so there were
probably no ill effects, we shouldn't configure the UART clock root in
each instance's configure call. This commit moves the UART root clock
configuration into each board's early setup routine. This should be
equivalent to what we were doing before.

The 'disable' term for the clock mux is misleading. This call actually
transitions the UART to use PLL3, not disable the clock to the UART.
(If the PLL is off, that's not obvious from this call.) The commit
refactors the CCM API to expose setters and getters.

Thread references to the CCM into other peripherals so that
the CCM clock gates don't need to refer to a static CCM instance.

Reduced size by a few hundred bytes.

They're now unused, so we can reduce LOCs.

Error messages won't be too helpful, but this component
shouldn't change much at this point.

Remove SNVS pad responsibility from IOMUXC module. Instead, directly
use the IomuxcSnvs type.