Board detection: derive target metadata from the catalog, explain detection decisions in place, and check probed pins per detector family

[ainyan03]

Commits

1. Derive board target metadata and detection expectations — targets.json no longer repeats chip, legacy board ID, board enum, descriptor name and C++ namespace; they come from the board catalog (two namespace exceptions remain explicit). Canonical board IDs are checked against the lgfx::board_t enum (octal, duplicate, conditional and macro definitions are rejected). Detect-class expectations are generated. A new host test runs every detector family with the production GPIO probes and fails if a pad is touched outside the detection snapshot.
2. Explain detection order, decisions and package facts where they are read — the generated detector lists, order constraints and GPIO power-hold IDs now carry board names and the existing reason from detector_order.json as comments, so { 154, 140 } reads as StampC6 before NanoC6 without a trip to boards.hpp. The common code states each decision where it is made (family ranking and scan stop, no_signature ending a session, which failures retry, the persistence condition, what rollback does not restore). def_flag_fallback / fallback_only() become def_flag_candidate_only / candidate_only(): these internal names were easy to confuse with the unrelated public fallback_board setting. A reason containing \ or ??/ is rejected, because a line splice would pull the next generated edge into the comment.
3. Check probed GPIO against each detector family's catalog pins — the snapshot check above uses the chip-wide union, so a probe moved to a pin another board already contributes went unnoticed. The host test now also requires the pads touched by signature() / confirm() to be inside the family's own catalog pins (computed by the same function as the firmware snapshot) and the C++ family members to match the catalog. It found the Atom family probing GPIO 2, 13, 21, 23, 27 and 34 without declaring them; the four Atom boards now declare the pins on their own path. Probe pins on a board without wiring_output never reach the firmware snapshot, so that combination is now rejected (E_TGT_DETECTION_WIRING).

Detection behavior is unchanged. With logging disabled, the allocated sections of M5GFX.cpp.o are identical to develop (ESP32 / ESP32-S3 / ESP32-P4 / ESP32-C6 with Arduino, ESP32-S3 with ESP-IDF 5.5), except for the renamed internal helper's section name. The firmware detection pin sets are unchanged.

Testing

• sh test/board_detect/run.sh (31 programs, 34 families), npm test / npm run check in extras/board_spec
• Each new check was confirmed to fail on a deliberate mismatch (NanoC6 probe pin changed in code only, in probe_pins only, an Atom touch pin removed from the catalog, a family member removed)
• Builds for the ESP32 family on the Arduino and ESP-IDF versions used in CI
• Fork CI passed