Fold the duplicated per-browser promotion loop into a helper

[dchaudhari7177]

Closes #335.

absorb_browser_analysis() takes a browser_analysis and folds the common collections plus the promotion loop, and both branches of run() become one line each.

The collections are folded by attribute name rather than by branching on the browser, as the issue asks:

BROWSER_COLLECTIONS = (
    "parsed_artifacts", "parsed_storage",
    "parsed_extension_data", "parsed_sync_data",
    "version", "preferences",
)

Chrome's parsed_extension_data and parsed_sync_data therefore need no special case, and Firefox growing one later needs no edit here.

On the installed_extensions decision

The issue asks whether that block stays outside the helper. It stays outside, and the reason turned out to be stronger than "it is different": moving it in is not behaviour-preserving.

I moved it in first and diffed the output. It produced one extra record:

{"source_long": "Chrome Extensions", "profile": "tests/fixtures\firefox",
 "ext_id": "uBlock0@raymondhill.net", "name": "uBlock Origin", ...,
 "data_type": "chrome:extension:installed"}

So Firefox does produce an installed_extensions block. It is parsed and then dropped, for exactly the same reason session_structure used to be — the promotion lives only in the Chrome branch. But promoting it as-is sends a Firefox profile's add-ons through the Chrome shape, and they come out labelled Chrome Extensions with data_type chrome:extension:installed, which is wrong for a Firefox profile.

That is a real finding and probably its own issue, but it is a behaviour change, and this issue is about removing duplication. Rather than smuggle it in, the block stays in the Chrome branch with a comment, and the helper's docstring records what is waiting there. Happy to open a separate issue for the Firefox side, or to fold it in here with the labelling fixed if you would rather have it in one go.

Two small notes on faithfulness

• The session_structure check stays hasattr, not a truthiness test. getattr(..., None) is not None would quietly change what an empty-but-present structure does, and this is the promotion that had already gone missing once.
• The promotion loop iterates browser_analysis.__dict__.items() instead of indexing __dict__[item] three times. Same values, same order, same broad except — I did not narrow it, since a promotion failure is deliberately non-fatal here.

Verification

python -m pytest tests/ -q --ignore=tests/corpus     248 passed, 2 skipped, 80 subtests passed
python -m pytest tests/test_corpus_e2e.py -q         3 passed, 2 skipped, 22 subtests passed

Output equivalence, which is the thing that actually matters for a refactor. Same fixtures, JSONL, before and after:

cmp before.jsonl after.jsonl   ->  identical, 377 records

The 120-line diff is 69 insertions / 51 deletions, most of which is the helper's docstring explaining the installed_extensions decision.