Add schema-free single-register loader

[bruno-f-cruz]

Summary

Introduces harp.data.read(source), the pyharp counterpart of harp-python's harp.io.read: read one register's binary .bin data into a pandas DataFrame, inferring the layout from the first frame. It is implemented on top of the unified harp.protocol register API instead of a hand-rolled byte parser.

Background — how harp-python loaded data

harp-python offered two layers:

• Low-level — harp.io.read(file, address=, dtype=, length=, epoch=, keep_type=): read a single register file, infer stride / timestamp / row count from the first message, and decode to the native numpy dtype. It hand-coded the Harp framing math (_dtypefrompayloadtype, stride/offset arithmetic, timestamp handling).
• High-level — harp.reader.create_reader(device) on top of harp.schema.read_schema + harp.model: parse a device.yml, generate per-register readers at runtime, and decode masks / enums / payloadSpec members from the schema. Files were resolved by the {device}_{address}.bin convention.

What this PR adds

• harp.data.read(source, *, timestamp=True, message_type=False) — inspects the first frame's header (payload-type byte, length byte, timestamp flag), builds the matching native register from harp.protocol (RegisterU16, RegisterS16Array(addr, length=n), …), and delegates to parse_to_dataframe → RegisterBase.parse_bulk. Decodes to native numpy types only (no enum / bit-mask decoding), matching io.read's behavior. source may be a path, bytes, or an open binary file.
• harp.protocol now exports decode_payload_type, encode_payload_type, and PayloadTypeInfo — the canonical payload-type-byte decoder that read relies on to infer the element type, size, and timestamp flag.