DiFluid R2: negative TDS decoded as ~655 % (concentration field is signed int16) #980

Problem

The native DiFluid R2 driver currently decodes the concentration/TDS field as unsigned. In difluid_r2_protocol.dart, _parseTds uses _readUint16(data, 1) / 100.0.

DiFluid's protocol document (protocolR2.md) shows the concentration field as a signed 16-bit value ×100. Its own Average example contains FF F9, which represents -0.07.

Negative readings around zero are valid. With the current native decoder, FF F9 becomes 655.29 % and is exposed that way through the Sensors API.

Expected

Decode the concentration/TDS field as signed int16, so FF F9 becomes -0.07.

Do not change refractive-index signedness as part of this issue. The signed concentration behavior is supported by the protocol example; there is not equivalent justification for treating the RI field as signed.

Scope

Keep this as a minimal correctness fix for the existing native fallback:

• decode concentration/TDS as signed int16;
• add a protocol regression test using a negative concentration packet;
• preserve existing positive TDS and RI decoding behavior.

No broader DiFluid R2 feature work should be added to the native implementation here.

Driver-runtime direction

Longer-term DiFluid R2 development is moving to the JS BLE Sensor driver runtime. #977 tracks that work, including native Average support.

The native R2 implementation remains useful as a fallback for now, so it should still decode valid readings correctly while that migration is in progress. Once the JS driver has feature parity, automated coverage and real-hardware validation, we can separately retire the native implementation.

Acceptance criteria

• FF F9 decodes to -0.07.
• Existing positive concentration decoding remains unchanged.
• Existing refractive-index decoding remains unchanged.
• A regression test covers the negative concentration case.

Hardware confirmation is welcome but is not required to establish this parser bug because the protocol example itself exercises the signed representation.