The scanning for "interesting lines" (function and class definition

headers shown on collapsed diff chunks) lived in the Differ, driven by

regex tables in diffviewer/filetypes.py. This change moves it into a new

diffviewer.interesting_lines module, in preparation for detecting

these lines with tree-sitter.
The new module hosts the regex tables, an InterestingLine type, and a

get_interesting_lines() entry point (currently regex-only).

DiffChunkGenerator.generate_chunks() now computes the interesting

lines for both file revisions directly, and _get_interesting_headers()

reads those instead of asking the differ.
The legacy Differ API keeps working for third parties:

Differ.add_interesting_line_regex() is deprecated for removal in Review

   Board 11.0 but still contributes results for now.
Differ.get_interesting_lines() runs the regex scanner, since no parsed

   tree is available at that layer. DiffChunkGenerator is the supported

   path going forward.
filetypes.py re-exports HEADER_REGEXES and HEADER_REGEX_ALIASES

   for backwards compatibility.

I've rewritten the interesting-lines tests as parameterized pytest cases

as part of this.

Ran unit tests.