Remove a useless pass separated from the next statement

[Eljees]

Fixes #65.

useless_pass_line_numbers recognises a leading pass only when the next real token sits on the row immediately after it:

start_row - 1 == last_pass_row

A blank line or a comment shifts that row, so the pass is never marked and survives:

def foo():
    pass

    a = 1

The pending pass is now carried across non-ATOM tokens (NL, COMMENT, INDENT, DEDENT) and resolved on the first ATOM token. The pending state is cleared on any ATOM rather than only a matching one — otherwise an atom at the same indentation in a later, unrelated block would be taken for the continuation of this one.

Measured

python -m pytest (excluding test_fuzz.py, which needs a corpus): 174 passed before and after.

source	before	after
pass, blank line, statement	not removed	removed
pass, comment, statement	not removed	removed
pass immediately before a statement	removed	removed
pass as the only statement in a function	kept	kept

The comment case is what shows this is line arithmetic rather than anything specific to blank lines.

Two things worth saying plainly:

• The blank line left behind where the pass was is not removed. autoflake doesn't manage blank lines and this change doesn't start; the issue text reads as if it should, so I'd rather name the difference than let it be discovered in review.
• On the already-working adjacent case the old code yielded the same line number twice ([2, 2]); it now yields it once. Nothing visible changes, since the numbers feed a membership test.

This touches useless_pass_line_numbers only. #349 rewrites filter_useless_pass further down the same file; the hunks don't meet.

AI-assisted (LLM used for drafting); the runs above are mine.