This patch enables output on the complimentary channels (TIMx_CHyN).
For timers 1 and 8, deadtime can also be inserted when the channels
transition. For the pyboard, TIM8_CH1/CH1N and TIM8_CH2/CH2N can
take advantage of this.

Found these by compiling stmhal with mp_uint_t of type uint32_t instead
of unsigned int.  This actually makes a difference to the code, but just
a curiosity.

Timers now have the following new features:
- can init freq using floating point; eg tim.init(freq=0.1)
- tim.source_freq() added to get freq of timer clock source
- tim.freq() added to get/set freq
- print(tim) now prints freq

Eg pyb.freq(120000000) sets the CPU to 120MHz.  The frequency can be set
at any point in the code, and can be changed as many times as you like.
Note that any active timers will need to be reconfigured after a freq
change.

Valid range is 24MHz to 168MHz (but not all freqs are supported).  The
code maintains a 48MHz clock for the USB at all times and it's possible
to change the frequency at a USB REPL and keep the REPL alive (well,
most of the time it stays, sometimes it resets the USB for some reason).
Note that USB does not work with pyb.freq of 24MHz.

Also improve precision of calculating PWM percent in integer mode.
Also update teensy with edge case fix.

Teensy doesn't need to worry about overflows since all of
its timers are only 16-bit.

For PWM, the pulse width needs to be able to vary from 0..period+1
(pulse-width == period+1 corresponds to 100% PWM)

I couldn't test the 0xffffffff cases since we can't currently get a
period that big in python. With a prescaler of 0, that corresponds
to a freq of 0.039 (i.e. cycle every 25.56 seconds), and we can't
set that using freq or period.

I also tested both stmhal and teensy with floats disabled, which
required a few other code changes to compile.

Also do the same for teensy timer code.

Fix stmhal and teensy print routines to report actual prescaler an period.
Fix teensy build to use soft-float
Add USE_ARDUINO_TOOLCHAIN option to teensy build

This allows to set the pulse width (for PWM mode) as a ratio relative to
the period of the timer.  Eg, 0.5 is a 50% duty cycle.  You can set the
ratio in the channel init, or using channel.pulse_width_ratio; the
latter can also read the pulse width as a ratio.

Addressing issue #50, still some way to go yet.

See discussion in issue #50.

This fixes #733.

Blanket wide to all .c and .h files.  Some files originating from ST are
difficult to deal with (license wise) so it was left out of those.

Also merged modpyb.h, modos.h, modstm.h and modtime.h in stmhal/.

Make include dependencies neater, and adheres to the coding convention
that headers should not include headers.

Simple but functional timer control.  More sophistication will
eventually be added, or for now just use direct register access :)

Also added pyb.freq() function to get MCU clock frequencies.

Internal flash used for the filesystem is now written (from the cache)
only after a 5s delay, or when a file is closed, or when the drive is
unmounted from the host.  This delay means that multiple writes can
accumulate in the cache, and leads to less writes to the flash, making
it last longer.

It's implemented by a high-priority interrupt that takes care of flash
erase and write, and flushing the cache.

This is still only an interim solution for the flash filesystem.  It
eventually needs to be replaced with something that uses less RAM for
the cache, something that can use more of the flash, and something that
does proper wear levelling.

Reads ADC values into a bytearray (or similar) at a fixed rate.  Needs a
better name and improved API.  Also fix up DAC dma function (which also
needs a better name and API).

This does not affect code size or performance when debugging turned off.

To address issue #420.

As per issue #257, servo is better on TIM5 because TIM2 is connected to
more GPIO.