Fix ADS1x15 init across library versions + differential mode

[msallin]

Summary

Four related changes to the ADS1115/ADS1015 read path. Builds on the work in #16 with a more portable pin-constant resolution and adds two follow-ups.

1. Sanitise sensor dict before persisting to conf — carries over the second half of Fix/ads1x15 pin constants #16. The read service mutates data[n]['object'] = AnalogIn(...) then str(i2c_sensors)-stringifies the whole thing, which produces <AnalogIn object at 0x...> and breaks the next eval() load. Build a serialisable copy at save time that strips runtime-only object and ranges keys. Channels stop disappearing from the UI after edits. Credits LeifYKlasson.
2. Robust pin-constant resolution — the previous attempt in Fix/ads1x15 pin constants #16 hardcoded Pin.A0..A3, which only exists in newer versions of adafruit-circuitpython-ads1x15. Pi installations with the older library (where Pin has no A0 attribute) hit AttributeError and the whole ADS1115 sensor goes red. Try three APIs in order — newer Pin.A0..A3, mid-era ads1x15.P0..P3 module-level, oldest ads1115.P0..P3 module-level — and use whichever resolves. Verified on a Pi where the original PR's fix raises AttributeError.
3. Fix getPaths2 publishing null for valid 0.0 readings — getPaths2 uses result = '' as a "no range matched" sentinel and emits null when result is falsy. But 0.0 is falsy too, so a legitimate range-mapped value of zero is silently turned into null. Concrete trigger: a 4-20 mA tank sender at empty tank (4 mA → raw 4800 → result 0.0). Use result != '' instead.
4. Optional differential-mode support for ADS1115/ADS1015 — the ADS1x15 hardware can read pairs (A0-A1, A0-A3, A1-A3, A2-A3) for common-mode noise rejection. The Adafruit driver already supports it via two-arg AnalogIn(ads, P0, P1). Expose through a new sensorSettings['diff'] field accepting "A0-A1" or "A0-A1, A2-A3". No GUI change required: the existing free-form sensor-settings dialog already accepts arbitrary keys. Backwards-compatible — when the setting is absent, behaviour is identical to single-ended.

Relationship to #16

This PR supersedes #16's pin-constant fix with a version-portable approach, and preserves #16's save-time sanitiser (with Co-Authored-By credit on that commit). If you'd prefer to land #16 first and then layer just the new bits on top, the differential-mode commit is the only one that depends on the pin work; the getPaths2 fix is independent and could be split out trivially.

Configuration example

Read a 4-20 mA tank sender across a 150 Ω shunt as a differential pair on A0/A1:

sensorSettings:
  gain = 1
  diff = A0-A1

A0 channel:
  SKkey       = tanks.fuel.0.currentLevel
  magnitudeSettings:
    range1 = 4800|28800 -> 0|1

A1/A2/A3 channels left unconfigured (A1 is consumed by the differential pair).

Test plan

• ADS1115 single-ended A0..A3 reading still works (no diff setting → identical code path as before).
• ADS1115 differential A0-A1 reading produces the same currentLevel as a standalone Python script using AnalogIn(ads, P0, P1).
• Pin resolution works on a Pi where adafruit_ads1x15.ads1x15.Pin.A0 does not exist (falls through to module-level P0..P3).
• Channels with magnitudeSettings mapping to 0.0 now publish 0, not null.
• Channels no longer disappear from the GUI after editing (no AnalogIn-in-conf corruption).

Open:

• ADS1015 single-ended A0..A3 reading (no hardware to test; code path is symmetric to ADS1115 and uses the same _A0.._A3 bindings).
• ADS1015 differential mode (same caveat).

[LeifYKlasson]

Good work! I missed these parts. If I understand it correctly, you have my changes in our PR plus your changes? So I can just close my PR? I'm completely ok with that, doesn't need the the credit for it :-), but of course ok if it is mentioned. Then I will also update my files and run them.

/Leif

[msallin]

@LeifYKlasson yes, it does include your changes. It was simplest to reapply them within one PR. Would be great when you can test it and report back if it works with your setup!