Add speed-conditioned dynamics limits

[vcharraut]

What

• Add a per-agent speed-conditioning coefficient.
• Scale classic and jerk dynamics’ forward-speed limits by that coefficient.
• Separate nominal forward, absolute forward, and backward speed constants.
• Normalize speed observations against the absolute 60 m/s limit.
• Keep IDM capped at the nominal 40 m/s limit.
• Update dynamics tests and notebook coefficient labels.

Why

Speed conditioning allows agents to learn different driving behaviors based on their assigned maximum speed. Separate constants clarify the distinction between the nominal dynamics limit, the maximum conditioned limit, and reverse-speed handling.

Notes

• Speed conditioning defaults to 1.0 when reward randomization is disabled.
• Randomized limits range from approximately 26.7 m/s to 60 m/s.

[Kait0]

otherwise looks good. Needs a healthy run to double check.

[Victorbares]

LGTM