Inside a list item, a sibling ordered marker at column <= the item's
marker_indent must end the current paragraph (CommonMark §5.2).
break_for_list_interrupt_after_inline_newline only ran the marker
lookahead when the leading whitespace contained a tab, so a non-tab
sibling ordered marker was missed whenever the parent's post-marker
space widened required_indent past the sibling's column.

Reproducer (biomejs#10374):

    1.  first
    2.  second
        1.   third
        2.   fourth

Inner item `1.   third` has marker_indent=4, required_indent=9. The
next line at indent 4 is a sibling, but `indent < required_indent` and
no tab in the leading whitespace, so the break check returned false and
`2.   fourth` plus its leading 4 spaces were swallowed as
MD_TEXTUAL_LITERALs inside the first inner item's paragraph.

The non-tab path now runs a narrow token lookahead that only flags
non-1 ordered markers — parse_inline_item_list's
textual_looks_like_list_marker already terminates on bullets and on
`1.`/`1)`, so widening to all marker shapes regressed
synthetic/nested-lists by ~10% without buying additional correctness.
The tab path keeps the original token-aware lookahead (biomejs#10333).

Verified:
- cargo test -p biome_markdown_parser (all 256 tests pass)
- cargo test -p biome_markdown_formatter (all 839 tests pass)
- cargo bench synthetic/nested-lists: cached +5%, uncached noise

Inside a list item, a sibling ordered marker at column <= the item's
marker_indent must end the current paragraph (CommonMark §5.2).
break_for_list_interrupt_after_inline_newline only ran the marker
lookahead when the leading whitespace contained a tab, so a non-tab
sibling ordered marker was missed whenever the parent's post-marker
space widened required_indent past the sibling's column.

Reproducer (biomejs#10374):

    1.  first
    2.  second
        1.   third
        2.   fourth

Inner item `1.   third` has marker_indent=4, required_indent=9. The
next line at indent 4 is a sibling, but `indent < required_indent` and
no tab in the leading whitespace, so the break check returned false and
`2.   fourth` plus its leading 4 spaces were swallowed as
MD_TEXTUAL_LITERALs inside the first inner item's paragraph.

The non-tab path now runs a narrow token lookahead that only flags
non-1 ordered markers — parse_inline_item_list's
textual_looks_like_list_marker already terminates on bullets and on
`1.`/`1)`, so widening to all marker shapes regressed
synthetic/nested-lists by ~10% without buying additional correctness.
The tab path keeps the original token-aware lookahead (biomejs#10333).

Verified:
- cargo test -p biome_markdown_parser (all 256 tests pass)
- cargo test -p biome_markdown_formatter (all 839 tests pass)
- cargo bench synthetic/nested-lists: cached +5%, uncached noise