Xiaomi S400: use 50 kHz for estimation equations

[miszobi]

Problem

S400BodyComposition (added in #1367 / #1376, cc: @DanyPM ) feeds the high-frequency
impedance band to the Sun 2003 (TBW) and Janssen 2000 (SMM) equations. Both were
derived on 50 kHz single-frequency BIA, and the S400's high-frequency band is
250 kHz.

For my setup this caused FFM and TBW values to be pushed out of range and suppressed.

Change

compute() already computes rL (the corrected low-frequency band) for the
Hanai ECW step. This PR feeds rL to Sun 2003 and Janssen 2000 as well, instead
of rH.

Testing

./gradlew :app:testDebugUnitTest — full suite green. The new
usesLowFrequencyForPrediction test fails without the one-line fix.

Generated and reasoned through with AI. Validated and edited by a human.

[oliexdev]

Before I go further into this one, a few questions — I'd like to understand the change before judging it.

1. After this change, does rHigh still feed any prediction? From reading compute() it looks like it's left with the Cole-Cole swap, the unreliableContact check and the empirical bone formula,
   which is why rH can be deleted. If that's right, TBW and ECW both come from the same resistance, and ICW = TBW - ECW is the difference of two equations on one input rather than something taken from
   the gap between the bands. How do you see that?

2. ecwTbwRatio moves from 0.42–0.45 to 0.46–0.48 across all three reference subjects. Still inside the 0.30–0.55 guard, but what's your read on those values?

3. Body fat rises by roughly 5 points for each reference subject (A 14.1 → 19.5 %, B 41.5 → 46.2 %, C 34.5 → 40.3 %), so every S400 user would see a step in their chart. Is that the intended magnitude?

4. How did you arrive at the new §7.1–7.3 expected values — by running the modified code, or checked against the specification document from S400 : implement dual impedance measurements #1367/Mi scale 2 dual impedance #1376? I recomputed A and B by hand and they match the
   new implementation exactly (A: TBW 45.07, FFM 61.57, SMM 33.84; B: TBW 26.79), which is why I'm asking: the test docstring still describes them as "the spec's reference subjects".

5. Can you say more about the case that triggered this — which values were suppressed, and what does the scale's own app report for the same weighing?

@DanyPM, since the pipeline and the spec are yours, your view on 1 and 4 would help a lot.