Adding Algo Registration + Connecting to Vis

[harshi-puli]

Summary

Replaces the manually-maintained ALGOS dict with decorator-based algorithm registration, wired through to vis.py.

• perception/tasks/registry.py (new): @register_perceiver(task=..., algo=...) decorator, plus get_perceiver, list_tasks, list_algos, discover_all(). Discovery walks perception.tasks and imports everything so decorators actually run — no shared file to hand-edit per new algorithm. Skips _archive/ and swallows any import failure (missing torch/kornia/tqdm, etc.) with a warning, so one broken/optional module can't block discovering the rest.
• Exisited Tasks: Current Task specifc class files were wrapped to be able to fit with the Task Perceiver process.
• vis.py: --algorithm/ALGOS replaced with --task/--algo via the registry. Unknown pairs now fail with a clear list of what's actually registered instead of a bare KeyError.
• Tests: registry behavior (lookup, duplicate keys, discovery) + per-task perceiver tests for slalom and path_marker.
• CI: temporarily disabled github workflow.

Verified

All three tasks run clean through the real vis.py CLI (synthetic + real footage). Full suite passes (20/20, excluding _archive's unrelated pre-existing breakage). Lint checked against pre-edit baseline on every touched file — no new findings.

• Verified that it also works on images and videos alike.

Test plan

• pytest --ignore=perception/tasks/_archive
• python -m perception.vis.vis --task <gate|slalom|path_marker> --algo <...> --data <path>
• Confirm an invalid --task/--algo prints the available list

[kvaikunthan]

ts tuff