fix(solis): stop trusting SolisCloud's slot metadata, and sweep the probe to a half amp

[springfall2008]

Fixes #5068. Follows #5309.

Problem

On the API-key route, SolisCloud's batch read returns a sysCommand definition with each register. Predbat capped slot currents at the smallest sysCommand.max across the six slots. That definition is a cloud UI entry for a list of model codes (productModel), and the cloud hands back entries for other models.

Live on an S5-EH1P5K-L (model 3104, 5 kW, one PureDrive LV pack at 51.2 V, CID 7224/7226 = 100 A):

Register	Advertised max	Applies to models	Held on the inverter
Slots 2–6 charge and discharge, discharge slot 1	60 A	f4,f7,f8,3101,3102,3109,3110,3121,3122,f5	100 A (slots 2–6)
Charge slot 1	62.5 A	3111,3140,3145	60 A (Predbat-capped)
CID 7224	70 A	long 31xx/5xxx list	100 A

3104 is in none of those lists. Predbat still capped slot 1 at 60 A (≈3.1 kW from the battery), so the customer saw exports of about 3.6 kW on a 5 kW inverter. max_export_power was 99999 (no limit). The 60 A entry does match model 3102, whose real 60 A ceiling #5309 measured by probing.

Change

• sysCommand metadata is ignored. The battery-side cap is the CID 7224/7226 register itself. The fix(solis): probe the slot current ceiling the inverter enforces but does not report #5309 probe starts there and measures what the slots really accept. Until it has run, the rated estimate is the fallback, as before.
• The probe sweep ends on a half amp. After 1 A up is refused, it tries 0.5 A up, so a ceiling such as 62.5 A is found exactly rather than rounded down to 62 A. Candidates are written as the inverter shows them (60, 62.5). This adds one write per direction to a probe.
• The limits log line no longer mentions metadata.

Testing

• test_slot_current_limits_inputs now covers the 5 kW case, using the real definitions and model lists from the inverter. The metadata does not lower the 100 A battery cap, and the unprobed cap is the 97 A rated estimate rather than 60 A.
• New test_slot_probe_finds_half_amp_ceiling: a 62.5 A ceiling is found as 62.5 A (62 kept, 63 refused, 62.5 kept).
• The existing probe sequences now include the half-amp step.
• Against main's solis.py, run test by test, all three fail: the metadata lowers the cap, 62.5 A is found as 62 A, and the sequence has no half step. All pass with the change.
• ./run_all --quick passes, and pre-commit passes on the changed files.

Notes

• Live on the 5 kW inverter with this branch:
  • Before the probe, the cap was the 97 A rated estimate (not the 60 A metadata), and slot_*_power_max/battery_rate_max read 4966 W.
  • The probe then learned 100 A both ways, from disabled slot 6.
  • Slot 6 already held 100 A, so that was taken as accepted without a new write. That is deliberate: a value the inverter already holds is treated as accepted, to avoid extra writes.
• tools/debug-journal.md's Solis row still describes the sysCommand minimum as a deliberate cap (the GH#5068 paragraph). It can be refreshed once this merges.

🤖 Generated with Claude Code