Inline synthio_sat16 and use ssat - #11621

* Take CircuitPython 11.0.0-alpha.2's integer mixer and inline synthio_sat16

audiomixer (11a329885, fc57c0f36; adafruit/circuitpython#11617): a voice is
scaled as (ai * mul) >> 15 where it was ai * (mul / 32767.0f) truncated toward
zero, and a centred pan is 32768 - panning where it was 32767 - panning. Taken
in upstream's words in src/audiomixer/Mixer.c, in the CPython extension's
scaling kernel and in src/cpython/audiomixer.py, whose float32 helpers go.

Every level below 1.0 moves by up to one LSB, and a small negative sample now
rounds down (-1 at level 0.5 stays -1). A voice at level 1.0 is a wire without
help, so the unity branch this port added in audiodsp#95 is deleted, and a
stereo voice at pan 0 no longer has its left channel one step below the right.
The CPython target's block-level pass-through at unity is keyed on Q15 32768
alone now: 32767 no longer multiplies by exactly 1.

synthio (cbbde1208; adafruit/circuitpython#11621): synthio_sat16 is inline in
src/synthio/__init__.h as upstream has it, with __ssat on cores that define
__ARM_FEATURE_SAT. The rounding is the same arithmetic without a branch, so no
sample changes.

mixer_q15_level_probe.py replaces mixer_level_precision_probe.py, whose
material was the float32-versus-float64 flips that no longer exist. It renders
the rails and the values either side of zero at four levels and five pans,
mono and stereo, and verify_dsp.py holds CircuitPython, MicroPython and CPython
to the same bytes. test_cpython_audiomixer_level.py and
test_cpython_mixer_scale.py are rewritten around the integer arithmetic.

* Move the oracle to CircuitPython 11.0.0-alpha.2 and re-capture two digests

CIRCUITPYTHON_ORACLE names d53b2277c, and tests/test_voice_ceiling_consistency.py
pins the sha256 of the 64-voice binary built from it (ee0cc93c...).

Two stored digests moved with the integer mixer, re-captured from the new
oracle's own output, which CPython and desktop MicroPython reproduce:

- streaming_component, streaming half: f3ffad8e... to 33b27a2f.... The mixer
  voices in streaming_component_probe.py are below unity.
- synthtools_acceptance: b6e3b06f... to 757d944a.... The lead, bend and bass
  lines moved; the other five lines are unchanged.

Every other stored digest held and reproduces on the new oracle:
midi_component, effects_component, biquad_component and mixdown_knee. The
oracle labels in effects_component and streaming_component follow the pin.

* Record the alpha.2 resync and retire "a voice at level 1.0 is a wire"

upstream-sync.md gets the new "last resync" entry (the alpha.1 one becomes the
one before it), upstream-diff.md says what upstream's integer mixer changed
and which digests moved, and the audiodsp#95 section becomes a record of a
difference upstream now shares, as the alpha.1 resync did for its five.

* Print the Q15 probe's levels by name

A float32 MicroPython build prints 100 / 127 with fewer digits than CPython,
so the probe's label line differed where the samples agreed.