schema(#293): get_signal_contact sense_ps_state per-PSU dict {psid: bool}

[AdamRickards]

Rank1 soft — get_signal_contact.sense_ps_state as a per-PSU dict (read side)

Related: #293

Anchor: 1.17 clerk trail (issue comment 5823364677). sense_ps_state is a per-PSU dict {psu_id: bool} on ssh, mops and snmp. The MIB table is hm2SigConPSEntry INDEX {hm2SigConID, hm2PSID}. When there are no rows the result is {}, never a bool and never None. Before this PR crude returned a scalar: mops True (last row won), ssh False (default), snmp collapsed by key_field.

Changes (YAML only; existing driver primitives, same recipe as archive signal_contact.yaml ps_state):

protocol	file	change
snmp	wire/diagnostic.yaml	index_fields + key_field: hm2SigConID (collapsed to one value per contact) → index: [group, inner]. Suffix SigConID.PSID gives {cid: {psid: bool}}, with HmEnabledStatus 1/2 → True/False through the existing matrix.
mops	wire/diagnostic.yaml	index_field: hm2SigConID + inner_index_field: _enumerate. The rows have no hm2PSID column, so PSID comes from the per-contact row order, as in 1.17 mops_hios and the archive. The unused index_fields list was dropped.
ssh	wire/ssh/diagnostic.yaml	new hm2sigconsensepsstate read on the existing show signal-contact {entity} all (iterate_from: [1, 2], which shares the iterate cache, so no extra show): parser: table, section: 1, key_column: 0, column: 1, tag: value_map, map: sigcon_sense_ssh (existing tokens monitored → true, default false).
schema	schemas/signal_contact.yaml	default sense_ps_state: false → {}, so a contact with no PS rows is honestly {}.

Offline checks (drivers + shaping, mocked I/O):

• snmp _gather_walks on {'1.1':1,'1.2':2,'2.1':1,'2.2':1} gives {'1': {'1': True, '2': False}, '2': {'1': True, '2': True}}. An empty walk gives {}.
• mops gather on 4 rows (SigConID + state only) gives the same nested dict. 0 rows gives {}.
• ssh _parse_response on a representative layout (title underline, then the Power Supply Status table, then the Intf Status table) gives {'1': 'monitored', '2': 'ignored'}, mapped to {'1': True, '2': False}. An empty or Error: response gives {}.
• _shape_table_output gives rows {'1': {'1': True, '2': False}, …}. With no PS data, every row gets {}.

No aggregation into a bool, no new primitive, no engine/driver edit, no invented tokens. validate_schemas.py --errors gives 46 | 45 | 0 errors | 1 warning, same as baseline.

Watch in proof / leftovers:

• The SSH layout is not device-captured. section: 1 comes from the archive and assumes a title underline before the PS table. If Test raw shows the PS table is separator 0, it's a one-token retune.
• The MOPS PSID-by-row-order is the 1.17 assumption the anchor flags. Test should confirm the keys match snmp's real PSID suffix.
• Write side is unchanged: set_signal_contact still has no per-PSU dict ingress, so that's out of scope for this read-side PR.
• sense_fan/sense_humidity/sense_stp_port_block None vs False, and the empty cli list (get_interfaces.read: SSH parse remainder (cli listed; n=1 vs mops 36) #226 class), are not touched here.

Proof: python3 tests/release_matrix.py --inspect --method get_signal_contact --trace --device