fix: track feature serialization and deserialization

[baszalmstra]

This fixes serialization and deserialization of track features. We serialized/deserialized this as either a string or a list of strings but in fact, the "standard" is to use a space-separated string. Deserialization is backward compatible with a list. But serialization will now always produce a space-separated string.

This only affects repodata and index.json format. Lock-files will still function the same way since they have their own parsing logic, and I think a list makes much more sense.

conda/conda-build#5608

[baszalmstra]

Supersedes #1038

[wolfv]

Indeed, I just tried with conda-build:

        "ttfeat-0.1.0-0.tar.bz2": {
            "build": "0",
            "build_number": 0,
            "depends": [],
            "md5": "f2dc829d0f36e4c125015591d07db7fc",
            "name": "ttfeat",
            "sha256": "f98a973b21a4943fc2872c7cb907d07e6219386509551d953b2e8062b1c37a01",
            "size": 2779,
            "subdir": "osx-arm64",
            "timestamp": 1738476674481,
            "track_features": "python tensorflow foobar",
            "version": "0.1.0"
        },

[wolfv]

LGTM! Should we call it serde::TrackFeatures instead of Features?

[baszalmstra]

LGTM! Should we call it serde::TrackFeatures instead of Features?

Its actually rhe same behavior for features. We dont parse that as a list yet because it would break the api but I thought Id already name it properly.

[wolfv]

OK! Features are really deprecated and broken so I don't think we'll ever need them but good to know :)