Fix pylint disable-next comments detaching from imports at function-body start

[Labib-Bin-Salam]

What's the issue?

Closes #2054.

When a function definition appears at the very start of a file (index 0), the first
comment inside the function body (at file-line index 1 or 2) was incorrectly flagged
by the in_top_comment mechanism. This caused it to be written directly to the output
stream before the import section was sorted, silently breaking
# pylint: disable-next=... annotations:

# Before fix – both comments float above all imports
def foo():
    # pylint: disable-next=no-name-in-module   ← detached
    # pylint: disable-next=no-name-in-module   ← detached
    from A import B
    from C import D

# After fix – each comment travels with its import
def foo():
    # pylint: disable-next=no-name-in-module
    from A import B
    # pylint: disable-next=no-name-in-module
    from C import D

Root cause

in_top_comment fired at indices 1–2 whenever no imports had been seen yet,
regardless of whether the preceding non-blank line was a genuine file header comment
(shebang / coding declaration) or actual code like def foo():.

Fix

Two targeted changes to isort/core.py:

1. top_comment_started flag – set only when a comment at index 0 opens a
   file-level header block. At indices 1–2, in_top_comment now additionally
   requires either top_comment_started (header continuation) or a # isort:
   directive prefix. Non-isort comments (e.g. # pylint:) inside a function body
   are no longer mistakenly promoted to top comments.

2. Indent inference from comment – when a non-top comment at index < 3 enters the
   import_section buffer (instead of being written directly), the current indent is
   inferred from its leading whitespace. This prevents a spurious indent-transition
   flush that would otherwise separate the comment from its following import before
   parse.file_contents can associate them.

Existing behaviour preserved

• File-level header comments (shebang, coding, licence) are still handled unchanged.
• Two top-level header comments separated by a blank line still both stay above imports.
• # isort: dont-add-import: and other # isort: directives at the top of a file
  continue to be treated as file-level control lines.
• Comments that are not at the very start of the file (e.g. block headers inside
  functions deeper in the file) retain existing behaviour – the indent-inference applies
  only to the first three file lines.

Tests

A new regression test test_pylint_disable_next_stays_with_import_issue_2054 covers:

• Two # pylint: disable-next annotations, each before a different import
• A single annotation where the annotated import sorts after an unannotated one
• An annotation preceded by a blank line inside the function body (fix still applies
  at index 2)

[DanielNoord]

Closing this and your other PRs until you engage with humans, see #2559

Creating PRs without engaging takes away valuable resources from maintainers.