This commit provides data structure and basic functions for the unified
control input handling.

…ructure

Move pitch and control event handling into separate functions and
expose the logic of prying apart the event code more clearly.

… copies

Maybe we can actually get rid of most of this copying. But for now,
this should work ... (untested so far, of course)

…iable

Processor selection is via a 3rd argument to midi():

- linear: normal linear control (default)
- accel_linear: control sends acceleration data; map to linear
- accel_cyclic: control sends acceleration data; cycle in interval [0, 1]

…election

This is only temporary to better track changes.

MIDI input now changes the register array directly. Note that this
means that writing to a control variable changes its value until
the next time it is updated by MIDI. Not sure if this is more
feature or bug.

This commit makes renderer_lock_patch, renderer_unlock_patch, and
renderer_get_patch produce meaningful resuls also when the renderer
is not running.

… patch

Not checking for this crashed FN when running the sampler without
rendering, e.g., in the audio dialog.

…mbers

A bit cleaner this way.

It's not generic.

accel_unbounded is (re)settable, as shown in T.fnp

Running a patch twice broke direct MIDI because pointers were updated
relative to the previous copy instead of the original patch. The commit
fixed the problem. The overall ugliness of the process remains, though.

…rect MIDI

The syntax is ugly: midi([chan], controller [, style])

Note that the comma after the channel is required, even if the
channel number is omitted. This would be difficult to do better
with the present syntax, but will get easier further down the
road.

…rmat

This commit changes the format from  CTRL (VALUE)  to either
CHAN.CTRL = VALUE  if the channel is the one currently selected, or
(CHAN.CTRL = VALUE)  if this is on a different channel.

The parentheses indicate that the traditional midiX variables are
not affected by this controller.

For compatibility, the number in the configuration is still zero-based.
We'll get rid of it eventually.

Also updated the example code in T.fnp

There are three interconnected changes in this commit:
- added quoted strings (TOK_STRING) to scanner,
- use them in the grammar, for midi_device,
- update T.fnp example

Also updated the example code in T.fnp

Thus made them "static" and removed them from the header.

For consistency with differential() and dt_diff.

For future sharing.

We use it also for range-to-toggle since this seems to be a better
approximation than requiring a zero-non-zero transition for each
toggle.

We use it also for differential-to-toggle, like in the range-to-button
case.

This just moves the ct->regs lookup across the API.