samd: Some improvements and bug fixes.

[robert-hh]

• Support MICROPY_HW_SOFTSPI_MIN_DELAY, increasing the SoftSPI baudrate to about 500kHz.
• Use __WFE() in the MICROPY_EVENT_POLL_HOOK, which reduces the wait time at the end of SPI read from up to 1 ms to a few µs.
• Avoid under-/overflow in the baudrate calculation. The underflow caused a baud rate setting of 48MHz resulting in a baud rate of ~100kHz.
• Fix UART IRQ flag setting and clearing. Without that fix, writing UART would disable the UART receive interrupt.
• Simplify machine_uart.c, by removing the call to uart_drain_rx_fifo() from machine_uart_any() and machine_uart_read(). It is not required, and may cause a race condition.
  method. Simplifies _boot.py as well
• Check the UART tx Pin assignment.
• Fix uart.deinit() and do not use the finaliser with the machine_uart object.
• Add a vref=n option to machine.ADC and machine.DAC. Some boards do not have the AREF input wired, which was used as default input. With the vref option, the reference source can be selected, e.g. to Vcc or to the internal reference.
• Support touch1200bps to start bootloader mode.

The first three changes are more inconveniences, which could be avoided in the Python scripts or "just" affect the timing. Nr. 4, 5, and 7 fix bugs, Nr. 6 is an additional test, Nr. 8 is for some boards a bug fix and in general an extension.

[dpgeorge]

I'm not sure why this works, because there are no SEV calls anywhere in this port, so WFE can never be woken.

I would prefer to use WFE rather than WFI because you have more control over when to wake the CPU, and it's easier to avoid race conditions. Using WFI you usually need to do this:

void wait(void) {
    for(;;) {
        disable_irqs;
        if (event/interrupt of interest occurred) {
            enable_irqs;
            return;
        }
        WFI;
        enable_irqs;
    }
}

WFE on its own can effectively replace the above function.

[robert-hh]

As far as I understood it, WFE also responds to a hardware interrupt, and maybe other hardware events, like from the . The difference here is, that with WFI the SPI unit wait for the next Systick interrupt after the last byte has been transferred, but using WFE it finishes immediately. That's why I replaced WFI by WFE.
Both instructions result in the same power consumption in REPL.

[dpgeorge]

OK, I guess the SAMD architecture has its own way of handling events/interrupts (I think the MCU can decide how to handle them, it's not part of the ARM Cortex spec, although I could be wrong there).

[dpgeorge]

Thank you! And thanks for splitting changes into neat commits.

Rebased and merged with very minor changes to formatting.

[robert-hh]

Thank you for merging.