fix: eliminate false fader events from ADC noise

[ArthurGibert]

Fixes false FaderChange events (and resulting MIDI CC spam in Jump mode) caused by ADC noise on static faders.

Root cause

Established by live probe diagnostics on hardware:

• Patch cables plugged into the jacks inject bursty noise into the jack traces, which couples into the fader wiper lines. A single trigger cable is enough; floating cables (far end on an unpowered module) are the worst carriers.
• Fader position amplifies the effect: at mid-travel the wiper source impedance is maximal (16-19 LSB measured wander) vs 1-2 LSB near the end stops. This is why the noise appeared position- and setup-dependent.
• The MAX11300 itself was ruled out: DAC write traffic, DAC step size, and ADC sweep range switching all measured clean.

Changes

• libfp/src/latch.rs: split jitter_tolerance into a tight emission dead zone (25 LSB) and a wider pickup_tolerance (50 LSB) for stay-latched hysteresis, with with_tolerances() constructor and characterization tests. Note this affects latch takeover behavior in Pickup/Scale modes for all apps, not just the fader task.
• faderpunk/src/tasks/max.rs: intra-dwell burst median — each fader read is the median of 5 back-to-back reads within one mux dwell. Rejects impulsive spikes with no added latency and no scan-rate loss.
• faderpunk/Cargo.toml: bump max11300 to 0.5.3, which skips redundant DAC data writes via a driver-side shadow register (feat: skip redundant DAC data writes via shadow registers max11300#27). An earlier revision of this branch implemented the dedup firmware-side in process_channel_values; that was reverted in favor of the driver, where &mut self serialization makes it race-free by construction.

An earlier revision also carried gated FADER_DIAG noise diagnostics in read_fader; removed to keep the hot loop lean (the measurement rig lives in this branch's history if needed for future noise sessions).

Validated live on an earlier revision of this branch: zero false emissions across all tested states, including the reproduced worst case (trigger cable patched, faders mid-travel, clock running), with fader feel confirmed unchanged. The checklist below re-validates on the final shape (firmware dedup replaced by the driver-side shadow, diagnostics removed).

Hardware test checklist

• Load a CV->MIDI layout in Jump takeover mode, patch a trigger/gate cable into an input, set all faders to mid-travel, leave faders untouched: no MIDI CC output / no value changes.
• Fader feel in all three takeover modes (Pickup/Jump/Scale), including slow fades and fast sweeps to the end stops.
• App-level latch takeover feel in Pickup and Scale modes (e.g. after a scene change): pickup engages at the expected proximity, no early/late latching (pickup_tolerance is now 50 LSB).
• CV outputs (e.g. LFO app) still update smoothly (DAC write dedup now in max11300 0.5.3).
• Layer switch (scene button) latching behaves as before.