Fix convert_osc corrupting text with multiple OSC sequences

[gaoflow]

Problem

convert_osc mangles text when the input contains more than one OSC sequence.
The loop uses re.finditer on the original string, but then mutates text
on every iteration. After the first removal the second match's start/end
offsets are stale — they point into the already-shortened string — so the slice
lands in the wrong place and leaves behind a partial OSC escape or drops
trailing content.

Minimal reproducer:

from io import StringIO
from unittest.mock import patch
from colorama.ansitowin32 import AnsiToWin32

with patch('colorama.ansitowin32.winterm', None):
    stream = AnsiToWin32(StringIO(), strip=True, convert=False)
    text = 'Hello\033]0;First\aWorld\033]2;Second\aEnd'
    print(repr(stream.convert_osc(text)))
# Before fix: 'HelloWorld\x1b]0;Second'   (garbage)
# After fix:  'HelloWorldEnd'

The same code path also raises AttributeError: 'NoneType' object has no attribute 'set_title' on non-Windows when winterm is None, as reported in
#407.

Fix

Separate the two concerns:

1. Side effects (calling winterm.set_title) — iterate all matches first,
   guarded by self.convert and winterm so non-Windows never touches winterm.
2. Text stripping — use re.sub('', text) once, which correctly removes
   all matches in a single pass.

Also tighten params[0] in '02' (substring test, matches '' and '02')
to params[0] in ('0', '2') (tuple membership).

Tests

Two new tests added:

• test_osc_multiple_sequences_stripped_correctly — asserts both OSC sequences
  are removed and surrounding text is preserved.
• test_osc_strips_without_crash_when_winterm_none — asserts no crash when
  winterm is None (non-Windows with strip=True, convert=False).

All 40 tests pass (14 skipped, Windows-only).