cpu/stm32: Restructure stm32 clk kconfig model

[MrKevinWeiss]

Contribution description

Currently nightlies are failing due to kconfig hash mismatches, ultimately due to clock tree configurations.
It is pretty hard to understand what is happening so this restructuring should be a first step to fixing the current issues (essentially not enough information is given to kconfig such as the HSE crystal speed, to make as good a choice for the clock settings are make).

This PR splits the Kconfig clock configurations for the stm32s from one file for all series, into many each series and any common combinations of series. This makes it more readable and matches what is done in make a bit closer.

As only the series that is applicable gets sourced, it also removes a lot of noise created by other series configurations being included.

Testing procedure

Murdock with nightlies enabled should have the same errors and passes as the latest nightlies.

Locally, all stms can be tested for an app with the following few lines:

Get a variable to hold all stm32 based boards

export TEST_BOARDS=$(FEATURES_REQUIRED=cpu_stm32 make -C tests/shell info-boards-supported --no-print-directory)

Overwrite the nightly flag to run all kconfig tests and binary comparisons with make

export NIGHTLY=1

Run through each stm based board for a single test, doing the kconfig comparison

for board in $TEST_BOARDS; do echo $board && /bin/bash -c "source .murdock; JOBS=4 compile tests/shell ${board}:gnu" | grep mismatch; done;

it should take maybe 10 to 20 mins, still faster than waiting for the CI.

Issues/PRs references

[MrKevinWeiss]

This is not meant to provide any fixes, we should have pretty much the same result as before. Only a restructuring, a follow-up PR will be provided that actually fixes the broken nightlies, but it is much easier to accomplish this (and test) with this one first.

[MrKevinWeiss]

Maybe ping @aabadie if he has an opinion on this (at least the idea as he was the one who did most of the work with the STM clock tree).

[riot-ci]

Murdock results

✔️ PASSED

b08c71d REMOVEME: Pretend it is nightly to do full tests

Success	Failures	Total	Runtime
1994	0	1994	08m:31s

Artifacts

• Documentation preview

This only reflects a subset of all builds from https://ci-prod.riot-os.org. Please refer to https://ci.riot-os.org for a complete build for now.

[MrKevinWeiss]

Hmm it seems that some CLOCK_PLL_M and CLOCK_PLL_N are hidden in some board files... I should do a search for that...

[MrKevinWeiss]

Also it looks like STM32_U5 is not modelled?

[MrKevinWeiss]

menuconfig looks a bit ugly, maybe this needs to all be hidden in a menu...

[MrKevinWeiss]

I wonder if breaking this up may create issues with potential speed-up methods in the future? If we just need one file to load rather then several depending on the CPU_FAM?

Any opinions @leandrolanzieri

[MrKevinWeiss]

probably will just be superseded by #18797 instead of dealing with rebase pain...