fix(antminer): read chain rate and board temps on newer stock firmware

[cryptographicturk]

Two parsing gaps on AntMiner stock firmware 86.48-2.0.0, both of which leave per-board fields None. Neither is model-specific — they affect the already-supported L9 exactly as they affect an L11.

The gaps

1. chain_rate{idx} is read as a string, but this generation emits a JSON number.

.get(&format!("chain_rate{idx}"))
.and_then(|v| v.as_str())          // firmware sends 6.790325248, not "6.790325248"

so board.hashrate is always None. The values are present in the RPC stats payload:

"chain_rate1": 6.790325248,
"chain_rate2": 6.791386112,
"chain_rate3": 6.754780672

2. temp_pcb{idx} does not exist on this generation.

The firmware splits the reading into inlet/outlet pairs instead:

"temp_in_pcb_1": "47",  "temp_out_pcb_1": "57",
"temp_in_pcb_2": "47",  "temp_out_pcb_2": "56",
"temp_in_pcb_3": "44",  "temp_out_pcb_3": "54"

so board.board_temperature is always None. This also leaves average_temperature empty, since it is derived from the per-board readings.

The change

• A small parse_f64_field helper that accepts either a JSON number or a quoted string, used for chain_rate{idx}.
• board_temperature falls back to temp_out_pcb_{idx} when the combined temp_pcb{idx} key is absent. The outlet reading is the meaningful one for thermal purposes.

Applied to both v2020 and v2023_07, which share this parser.

Measured

Against live hardware, per board:

before:  temp=None      hashrate=None
after:   temp=53C       hashrate=6.80 GH/s

average_temperature populates as a consequence. Verified on both an L9 and an L11 running this firmware.

Verification

• cargo fmt --all -- --check — clean
• cargo check --workspace — clean
• Live check against multiple L9 and L11 units on a production farm

Note

Board temperature could reasonably map to the 4-field temperature model added in 0.7.0 rather than collapsing to the outlet reading — temp_in_pcb_* / temp_out_pcb_* and temp_in_chip_* / temp_out_chip_* are all present on this firmware. I kept this change minimal and behaviour-compatible; happy to extend it if you would prefer the fuller mapping.