…HTML (#9…"

This reverts commit 0e0dbdf.

…egression

- Quoted lazy continuation: a `>` quote prefix on a continuation line no
  longer terminates the surrounding paragraph. The terminating-block
  check is now restricted to unprefixed continuation lines, matching
  CommonMark.
- Split `ordered_marker_non_one_no_interrupt.md` into per-case fixtures
  (`ordered_marker_non_one_in_bullet_item`,
  `ordered_marker_one_starts_sublist_in_bullet`,
  `ordered_marker_non_one_top_level`) per @ematipico's review.
- Restore the entity-whitespace fix: numeric character references that
  decode to whitespace (e.g. `&biomejs#9;` → `\t`) are now preserved across
  the paragraph close-time trim by tracking entity output ranges
  out-of-band, restoring 100% CommonMark conformance (652/652).

…ion lines

`strip_paragraph_indent` ran `trim_start()` on each continuation line and
stripped whitespace that originated from a decoded numeric entity (e.g.
`&biomejs#9;` → `\t`, ` ` → ` `), even though the paragraph buffer already
tracks those bytes in `entity_ranges`. Make the helper entity-aware: it
now accepts the buffer's entity ranges plus the trimmed slice's offset
within the original buffer, and halts the per-line whitespace trim on the
first byte that lives inside any preserved range. Structural literal
indentation is still stripped as before.

Regression tests cover entity-sourced tab, entity-sourced space, and the
mixed case where literal leading indent precedes an entity-sourced tab.

…egression

- Quoted lazy continuation: a `>` quote prefix on a continuation line no
  longer terminates the surrounding paragraph. The terminating-block
  check is now restricted to unprefixed continuation lines, matching
  CommonMark.
- Split `ordered_marker_non_one_no_interrupt.md` into per-case fixtures
  (`ordered_marker_non_one_in_bullet_item`,
  `ordered_marker_one_starts_sublist_in_bullet`,
  `ordered_marker_non_one_top_level`) per @ematipico's review.
- Restore the entity-whitespace fix: numeric character references that
  decode to whitespace (e.g. `&biomejs#9;` → `\t`) are now preserved across
  the paragraph close-time trim by tracking entity output ranges
  out-of-band, restoring 100% CommonMark conformance (652/652).

…ion lines

`strip_paragraph_indent` ran `trim_start()` on each continuation line and
stripped whitespace that originated from a decoded numeric entity (e.g.
`&biomejs#9;` → `\t`, ` ` → ` `), even though the paragraph buffer already
tracks those bytes in `entity_ranges`. Make the helper entity-aware: it
now accepts the buffer's entity ranges plus the trimmed slice's offset
within the original buffer, and halts the per-line whitespace trim on the
first byte that lives inside any preserved range. Structural literal
indentation is still stripped as before.

Regression tests cover entity-sourced tab, entity-sourced space, and the
mixed case where literal leading indent precedes an entity-sourced tab.