cpu/sam0_common: flashpage: attempts at bugfixing

[benpicco]

Contribution description

This 'fixes' what I thought would be the cause for #14929

• wait for commands to be done before proceeding
• disable cache to avoid triggering NVM Read Corruption errata

But this just happens to push the issue away temporariely.
If you replace the interrupt based wait_for_cmd() with a polling

static inline void wait_for_cmd(void)
{
   while (_NVMCTRL->INTFLAG.reg == 0) {}
    _NVMCTRL->INTFLAG.reg = 0xff;
}

or use a different C library or compiler, the bug is back again.

Testing procedure

Flash tests/mtd_flashpage on same54-xpro.
The test will now succeed.
But this is a fluke: try again with PICOLIBC=1 and it will again fail.
(The test would succeed on master with PICOLIBC=1, but fail with newlib. The C library is entirely unrelated to the test.)

Issues/PRs references

attempts to fix #14929 (but no sucess so far)