BUG: fix IndexError when usecols selects a column not at position 0

[Himesh-rupchandani]

Closes GH#67375.

What

read_csv(usecols=...) with low_memory=True raised IndexError when a
usecols-selected column was at a position other than 0 and different chunks
inferred mixed dtypes for that column.

Root cause

In _concatenate_chunks, chunk keys are raw column positions (e.g. 50), but
column_names is already filtered by usecols, so column_names[name]
indexed by the raw position and went out of range. It only worked when the
column happened to be at position 0.

Fix

Build a position->name mapping from the sorted chunk keys (which align with
the ascending-position order of column_names) and look up by position
instead of positional index.

Test

Adds test_concatenate_chunks_usecols_mixed_dtype_nonzero_pos in
test_concatenate_chunks.py. It fails before the fix (IndexError) and passes
after.

AI assistance

I used an AI assistant to reproduce the bug, locate the cause, and draft the
fix and test; I reviewed all changed lines and verified the behaviour locally.

[Himesh-rupchandani]

Hi, I've reproduced this and prepared a fix + regression test. PR is up.

Summary: in _concatenate_chunks the chunk keys are raw column positions but
column_names is already filtered by usecols, so column_names[name] indexes
out of range for a selected column at position > 0. Fix: map position->name
from the sorted chunk keys.