Unify impedance CSV output across streaming and non-streaming paths

[namthor9]

Summary

Makes synapsectl query and synapsectl query --stream produce identical impedance CSVs, fixes pre-existing format discrepancies, and adds a peripheral metadata header line.

Unified format (both paths):

Peripheral: NYX1-512
Electrode ID,Magnitude (Ohms),Phase (degrees),Status
8,5523878.0,-83.439453125,1
10,5462193.0,-84.0098648071289,1

Changes

• New shared module synapse/cli/impedance_csv.py — single source of truth for the header, columns, row format, and peripheral-name resolution. Both paths use it, so output is byte-identical (LF line endings forced).
• New Peripheral: <name> header line. The name is resolved from device.info(), preferring the peripheral_id named in the query, falling back to the broadband (recording) source, then the first peripheral, else Unknown.
• Fixed streaming discrepancies. The streaming path previously wrote a different header (Electrode ID,Magnitude,Phase), omitted the (Ohms)/(degrees) units, and had no Status column. It now matches the non-streaming format.
• Streaming now records failed measurements with Status=0 (success = 1). Previously failed electrodes were dropped from the CSV entirely.

Notes

• Peripheral name comes from device.info(), not the impedance response. For a single-recording-device setup this is exact; with multiple broadband sources and no peripheral_id in the query it picks the first. Returning the peripheral in the response itself would require a synapse-api proto change (out of scope here).
• peripheral_id lookup uses getattr(..., "peripheral_id", 0) so it's robust to older protos that lack the field on ImpedanceQuery.

Testing

• py_compile passes on all three files. Generated *_pb2.py are gitignored, so full runtime exercise requires a built environment / device.

[calvinleng-science]

Make info() resolution consistent across both paths. The streaming case resolves the peripheral name before stream_query(), and the non-streaming case resolves it after.

We should make both resolve the peripheral name prior to the query, as if something happens to the probe to make it unenumerate after the query, we will not be able to correctly resolve the peripheral name. Just more robust this way.