fix(feature): avoid IndexError when parsing a dtype with a trailing dot

[devteamaegis]

What's broken

parse_nested_brackets crashes with IndexError: string index out of range on any dtype segment ending in a dot, e.g. "bionty.". It's reachable from the public parse_dtype via a malformed categorical dtype:

parse_dtype("cat[bionty.]")

  File ".../lamindb/models/feature.py", line 300, in parse_nested_brackets
    if len(parts) == 2 and parts[1][0].isupper():
IndexError: string index out of range

Why it happens

After dtype_str.split("."), the code checks len(parts) == 2 but then reads parts[1][0] without checking that parts[1] is non-empty — a trailing dot makes it "".

Fix

Guard parts[1] != "" before indexing. The malformed input now falls through to the bare-registry branch, so parse_dtype("cat[bionty.]") raises a clear ValidationError ("invalid dtype") instead of an opaque IndexError.

Test

test_parse_nested_brackets_trailing_dot checks "bionty." no longer crashes and parse_dtype("cat[bionty.]") raises ValidationError.

[falexwolf]

Thank you very much for the contribution! Will merge a version of this.

Given you ran into this: Have you considered passing Python objects instead of strings?

[falexwolf]

I think what we should probably do is reason through a few more cases in which strings that a user passes could raise opaque errors.

This fix is good but it also feels a bit ad hoc; also testing strategy.

Thank you anyway, we'll mull over this for a bit more.