[bdbarnett]

A jump of a level or a Mix now walks over 20 ms instead of stepping, in the nodes that carry them. This is the first stage of the control-smoothing plan audioeffects follows: Mix, levels and depths first, then filter corners, then crossfaded patch changes.

What walks:

• audiomixer.Mixer gets a level_ramp property, a number of frames a voice's level or pan move takes, in a straight line, instead of waiting for the voice's next zero crossing. At 0, the default, the Mixer renders CircuitPython's bytes. CircuitPython's zero-crossing wait stops a single voice stepping mid-cycle, but a dry and a wet voice cross zero at different moments, and a voice that never crosses zero waits for the end of a stretch and then lands at once. The new test's crossfade into a voice that stays positive steps 14.6 times the tone's own largest step without the ramp and 0.96 times with it. How it walks is in docs/how-it-works/upstream-diff.md, "A level ramp on Mixer voices".
• The Waveshaper's mix and post_gain, the Biquad's and the AllPass's mix, the Dynamics node's makeup and the FeedbackDelay's mix walk a move over 20 ms (src/shared/audiodsp_level_ramp.h). These nodes are this package's own, so nothing upstream changes. A node's first value lands at once, so a fresh node renders as it did, and a reset clears a walk in progress.
• audioroute.Port.play(source, after=frames) keeps playing what the port plays until that many frames have gone, then switches at the start of the next pull. An effect uses it to walk its graph to the dry before handing over the source, so Mix 0 is still the source byte for byte without a cut on the way there. Pointing a port at the source it already plays is no longer a new start, so it no longer cuts a switch short.

The audioroute stub declares Port.play(after=) too, so a type checker accepts a deferred switch; checked with mypy over the stub and tests/test_stubs_match_modules.py.

What I ran: the CPython tests for every module this touches (audiobiquad, audioecho, audiodynamics, audioshaper, audioroute, audiograph nodes, the mixer level ramp, the new node ramp tests, binding parity, free threading and the API test: 441 tests, all passing); the five CI gates; and verify_dsp.py with CPython and a unix MicroPython built from this branch, 65 comparisons and 0 failures, including the new node_level_ramp_probe.py and mixer_level_ramp_probe.py. delay_lifecycle_probe.py held the FeedbackDelay's output at Mix 0 to the input from the first frame; it now holds it from where the 20 ms walk lands and checks the walk is there. In audioeffects, renders of the 15 classes that build on these nodes step at most 1.09 times the tone's own steepest step when Mix or a level jumps, where they stepped up to 62 times before.

Not verified: the three-way run with the CircuitPython oracle (its pinned binary predates a keyword filter_f32_frames_probe.py uses and stops the run there, unrelated to this change), and no board build.