fix(adapters): aggregate linked members at the range a lock pins

[jvdd]

Fixes the linked case of #82. The unlinked case needs a design and is not in this PR.

Change

• fvLockAxisGroup (toolbar.js) compares the range each group member shows with the range of the clicked figure. If a member differs, the lock writes the group range as the viewport of the group with fvWriteViewport.
• Two ranges are equal when each bound differs by less than 1e-9 of the span (fvSameRange). Figures linked on one column rarely show bit-identical ranges: a line shows [3.1e-13, 9999.0000000001] where a histogram shows [0, 9999]. Date strings keep the exact compare.
• fvOnToggleAxisLocks commits the written keys with fvCommitViewportChange. fvOnLockAllAxes sends one commit for all groups.
• A group whose members already show one range stays display-only: no viewport write, no request.
• Architecture.md and docs/guides/cross-filtering.md describe the new behavior. Architecture.md also states that an unlock writes no viewport: a group whose members differed keeps its viewport keys, and a display-only group has none.

Impact

• Before, a linked member that had autoranged to other data showed the locked window over data that the server had aggregated for its own full range. Measured: A is a line on 0..999, B a line on 0..9999, linked on x, n_points=200. After a lock on A, B showed 20 of its 200 points inside the window. Now all 200 are inside.
• A lock on a line and a histogram of one column sends no request. With an exact compare, it sent one.
• After an unlock, the viewport keys stay, so display, state and aggregation keep agreeing. A double-click or reset returns to the full range, as after a zoom.

Known limit

The validator requires equal viewport values in a link group, so when members differ, the clicked figure gets the group viewport too and is re-aggregated at the range it shows. A binned trace then bins on the zoomed lattice. On the weekday hist2d (y 1..7, y_bins=7) the rows go from 7 to 8, with one empty row. A manual zoom to the same range gives the same grid on main. #124 tracks the server fix.

Tests

• test_lock_aggregates_a_member_at_the_range_it_shows: counts B's points inside its window after a lock on A. Fails before the fix (20 of 200).
• test_lock_all_sends_one_request_for_every_group: two groups whose members differ give one request with all four keys. Fails before the fix (no request).
• test_lock_on_members_that_show_one_range_posts_nothing: guards the display-only rule.
• test_lock_on_a_line_and_histogram_of_one_column_posts_nothing: fails with the exact compare (one request posted).
• Tree merged with main, pytest tests flexviz_polars/tests: 2425 passed, 2 skipped.
• Tree merged with main, pytest -m browser tests: 334 passed, 12 skipped.

[codspeed]

Merging this PR will not alter performance

✅ 31 untouched benchmarks

---

_{Comparing fix/linked-lock-viewport (4387721) with main (6213d31)}