Flash sector erase does not work properly on STM32U5

[williams-one]

I'm testing flash sector erase on a STM32U5G and it doesn't work properly in all conditions.
I have tried to analyze the issue and here are my findings. If you want, I'm more than happy to work on a patch 😄

I think that this could be a summary of all the concurrent causes.

First, all U5 are dual bank but there is no selection of bank register (BKER) in blocking_erase_sector

embassy/embassy-stm32/src/flash/u5.rs

Lines 69 to 79 in fb8e40b

	pub(crate) unsafe fn blocking_erase_sector(sector: &FlashSector) -> Result<(), Error> {
	#[cfg(feature = "trustzone-secure")]
	pac::FLASH.seccr().modify(|w| {
	w.set_per(pac::flash::vals::SeccrPer::B_0X1);
	w.set_pnb(sector.index_in_bank)
	});
	#[cfg(not(feature = "trustzone-secure"))]
	pac::FLASH.nscr().modify(|w| {
	w.set_per(pac::flash::vals::NscrPer::B_0X1);
	w.set_pnb(sector.index_in_bank)
	});

Then there is a problem with page size. According to ST specs

But the erase_size in the metapac is set to 16384, while I think it should be 8192 to match the size of the page
https://github.com/embassy-rs/stm32-data-generated/blob/5373e57121b8300a3c2f03fb0408d857f6622607/stm32-metapac/src/chips/stm32u5g9zj/metadata.rs#L8-L18

This has been fixed for some variants of the U5 family but not for all
https://github.com/embassy-rs/stm32-data/blob/55b491ee982e52f80a21276a0ccbde4907982b5d/stm32-data-gen/src/memory.rs#L352-L353

Finally there is also another problem in the function for setting the PNB (page number) register here
https://github.com/embassy-rs/stm32-data-generated/blob/5373e57121b8300a3c2f03fb0408d857f6622607/stm32-metapac/src/peripherals/flash_u5.rs#L506-L516

According to ST specs PNB should be 8 bit, so I think the masking is wrong. It should probably 0xFF instead of 0x7F

This is not a problem for devices which has 2MB of flash or less since the maximum page index is 127, but it is a problem for devices with 4MB of flash which have a maximum page index of 255