Generalize decoder implementation

[mojomex]

Description

Currently, similar models from the same vendor share very little of their decoder code despite their decoders working largely the same. This issue calls for a generalized decoder (at least per vendor) that is only specialized where necessary.

Purpose

• De-duplicated code is much more maintainable/understandable
• Optimizations/bugfixes for one lidar decoder would automatically apply to all decoders

Details

For Hesai lidar packets:

• The pre-header and header fields are the same format (except for Pandar-64 (short header) and Pandar-40P (no header))
• The block (≈ lidar column) format is identical except for an optional fourth (mostly unused) byte
• The body (group of blocks) format is the same except for number of blocks, fine azimuth and checksum fields
• Formats in the tail are wildly different. However, only a few fields are used by Nebula and those do have the same format
• Return modes can be handled in a generalized way (format is consistent across models)

The overall algorithm of decoding the packets is basically identical (parameterization is of course model-dependent).

Possible approaches

Implement the decoding algorithm fully generically, using specific model definitions (as they are already largely present) for parameterization. The model definitions should re-use structs for headers/blocks whereever possible. If the structs are packed, the memcpy-based parser from pandar_128_e4x_decoder can be used for all models. Accessor functions to the specific struct locations can then provide the necessary abstraction.