fix(core): reject transcript timestamp ranges

[phernandez]

Why

Transcript time ranges such as [24:33.098 - 24:41.260] still matched Basic Memory's bracketed observation syntax. This is the remaining range-shaped case from #1219/#1239 and causes one junk observation category per transcript line.

Closes #1270.

What Changed

• Recognize one clock value or a spaced range of two clock values as transcript timecode syntax.
• Keep those prefixes as ordinary Markdown content instead of observation categories.
• Preserve semantic categories that merely contain numeric or time-like text, including [10:30am - 11:30am].

Implementation Details

The existing narrow clock grammar is factored into one reusable regex fragment and composed into an optional spaced-range form. It still accepts only MM:SS/HH:MM:SS-shaped values with optional dot or comma fractional seconds; it does not broaden the exclusion to arbitrary numeric categories.

Testing

Automated

• uv run pytest -q --no-cov tests/markdown/test_observation_edge_cases.py: 13 passed
• just fast-check: passed Ruff fix/check, formatting, and ty type checking
• git diff --check: passed

Manual

• Not applicable; parser behavior is covered directly by regression cases.

Risks / Follow-ups

The matcher intentionally requires whitespace around the hyphen. Other transcript range separators can be added from concrete reports without guessing at semantic category syntax.