fix(frontmatter): handle quote characters inside unquoted inline-array values

[hyun-oho]

Problem

The frontmatter dialect parses inline arrays (links: [a, b, c]) with a
single-pass state machine in _split_items. The quote-tracking loop
treats any quote character (' or ") as a toggle into/out of
quoted mode, even when the quote appears mid-value rather than at the
value boundary.

An unquoted value containing a quote character — it's for example —
enters quote mode on the apostrophe and never exits. Every subsequent
comma is then swallowed as part of the quoted string, so the array
collapses into a single broken item.

>>> frontmatter.parse("---\nlinks: [it's, done]\n---\nbody\n")
{'links': ['"it\'s, done"']}    # wrong: 1 item, includes the comma

The render path always wraps list items in quotes, so the round-trip
via render → parse happens to work. The bug surfaces when a front-
matter file is edited by hand (which the project encourages — the
files are the source of truth) and an unquoted value contains an
apostrophe or a double-quote character.

Fix

Track whether the cursor is at the start of a new value (right after
the opening bracket, a comma, or leading whitespace). Only enter quote
mode when a quote character appears at that position. A quote in the
middle of an unquoted value is now treated as a regular character.

>>> frontmatter.parse("---\nlinks: [it's, done]\n---\nbody\n")
{'links': ["it's", "done"]}     # correct: 2 items

Testing

Added tests/unit/test_frontmatter.py covering the parse / render
round-trip, the unquoted-apostrophe case, a quoted value that contains
the other quote type, and the existing scalar and empty-array paths.

$ python3 -m pytest tests/unit/test_frontmatter.py -v
10 passed in 0.17s