Race condition in ow.write() / ow.write_bytes() with power=1

[devsaurus]

There's a race for ow.write() / ow.write_bytes() with power enabled between the driver code to activate the strong output driver and any random interrupt. Looking at `onewire_write_bit() for the 0 case:

• write '0' slot generation is protected against interrupts
• unprotected phase begins and contains some µs delay plus the return back to onewire_write()
• interrupts are masked again and the output driver is activated in line 175

If an interrupt occurs during the unprotected phase or a previously pending interrupt gets scheduled, the delay between the last bit's write slot and the activation of the strong output driver will be extended by some 10s of microseconds. This is a potential stability issue for ds18b20 et.al. sensors in parasite power mode which dictate T_SPON(max) to be 10 µs.

I suspect that the strong driver has to be activated at the end of the last bit, still within the phase where interrupts are masked by onewire_write_bit().