Add byte-level CANShield IDS module

[shubham-899]

Extends CANShield to operate on raw CAN payload bytes without DBC files. Adds per-CAN-ID anomaly scoring. Implements ids.base.IDS interface. Auto-discovered via ids/init.py.

[shubham-899]

Summary
Adds a byte-level extension of CANShield as a CAN-Rakshak IDS module.

CANShield operates on DBC-decoded signals, requiring proprietary decoding
files unavailable for most vehicles. This module extends it to operate
directly on raw CAN payload bytes, removing the DBC dependency, and adds
per-CAN-ID anomaly scoring to overcome feature dilution on single-byte attacks.

Implementation

• New file: ids/canshield_byte.py
• Implements the ids.base.IDS interface (train/test/predict/save/load)
• Reads the framework-standard CSV (timestamp, can_id, dlc, b1..b8, label)
• Auto-discovered via ids/__init__.py (no manual registration)
• Use by setting MODEL_NAME = "CANShieldByte_demo" in src/config.py

Method

• Forward-filled byte matrix over top-N CAN IDs (no DBC decoding)
• Correlation-based column clustering
• Multi-period (T=1,5,10) CNN autoencoder, trained on normal traffic only
• Per-CAN-ID scoring: max over IDs of fraction-flagged bytes

Unsupervised — trains on normal data only, no attack labels needed.
No new dependencies beyond the framework's existing ones.