fix(utils): strip the _qlib_ prefix in fname_to_code, not characters

[he-yufeng]

Description

fname_to_code() is meant to reverse code_to_fname(), which prefixes reserved Windows device names (CON, PRN, AUX, NUL, COM0–COM9, LPT0–LPT9) with _qlib_ so they can be used as directory names. The reverse used fname.lstrip("_qlib_"), but str.lstrip removes any leading characters contained in its argument, not the prefix as a whole.

exists_qlib_data() lowercases each feature directory name before calling fname_to_code():

code_names = set(map(lambda x: fname_to_code(x.name.lower()), features_dir.iterdir()))

So LPT1 is stored as _qlib_LPT1, lowercased to _qlib_lpt1, and then lstrip("_qlib_") strips the leading l too (it is one of the prefix characters), yielding pt1 instead of lpt1. The data-existence check then fails to match such a stock. Any name whose body begins with a character in {_, q, l, i, b} is corrupted (e.g. fname_to_code("_qlib_lll") returns "").

The fix removes the prefix by slicing, so the original code is recovered intact.

Verification

Before:

>>> fname_to_code("_qlib_lll")
''
>>> fname_to_code(code_to_fname("LPT1").lower())
'pt1'

After:

>>> fname_to_code("_qlib_lll")
'lll'
>>> fname_to_code(code_to_fname("LPT1").lower())
'lpt1'

Added FileNameUtils.test_fname_code_round_trip in tests/misc/test_utils.py, covering the reserved device names (via the lowercased path exists_qlib_data uses), a prefix-only body, and a plain code. black -l 120 --check passes on the changed files.