Grid: resolve named lines once (before the grid size estimate)

[nicoburns]

Objective

compute_grid_size_estimate treated every named-line placement as auto (into_origin_zero_ignoring_named), but its negative implicit track count is used as exact, and its positive count as a lower bound. Items placed with named lines therefore created implicit tracks that nothing occupies:

• grid-column: foo / 1 (unknown foo, resolves to 1 / 4 in a 2-column grid) was estimated as auto / 1, adding an implicit column before the grid.
• grid-row: span 5 / last was estimated as a fully auto-placed span of 5, adding implicit rows after the grid as well as the ones the item really needs before it.

The estimate now works on each child's grid-row/grid-column with named lines already resolved. To avoid resolving names twice (once for the estimate, once in place_grid_items), name resolution is moved ahead of both: compute_grid_layout resolves every in-flow child's named lines exactly once and hands the results to both the estimate and place_grid_items.

This fixes 2 WPT subtests in Blitz (grid-placement-unknown-named-grid-line-001.html, grid-placement-negative-position-resolution-001.html), with no regressions in css/css-grid.

Blitz PR: DioxusLabs/blitz#1072

Context

compute_grid_layout builds items and placements: Vec<ItemPlacement> (InBothAbsAxis<Line<OriginZeroGridPlacement>> — the grid-row/grid-column styles with named lines resolved, in origin-zero coordinates; previously built inside place_grid_items step 0) in one walk over the in-flow children, then:

compute_grid_size_estimate(explicit_col_count, explicit_row_count, &placements)
place_grid_items(&mut cell_occupancy_matrix, &mut items, &placements, grid_auto_flow)

• compute_grid_size_estimate / get_known_child_positions take &[ItemPlacement] instead of a style iterator (no generics, no resolver); child_min_line_max_line_span takes the Line<OriginZeroGridPlacement> and is otherwise unchanged, so negative implicit tracks are still pre-sized by the estimate.
• place_grid_items loses step 0 and the children_iter/align_items/justify_items/named_line_resolver parameters; steps 1–4 are unchanged.
• ItemPlacement stays in placement.rs, now pub(super) and re-exported through the grid module for implicit_grid.rs.
• util::test_helpers gains resolve_named_lines (resolve test styles the same way) and grid_items (build test GridItems), used by the implicit_grid/placement unit tests.

Memory is unchanged (placements was already allocated by place_grid_items; items is reserved from child_count as before). main walked the children twice (estimate, then placement), calling grid_row()/grid_column() — which clone the placement — on each walk; the single walk is measurably faster.

Benchmarks vs the merge-base d7cb4642 (release, interleaved rounds pinned to one core, min of per-round Criterion medians; 15 rounds for named, 5 for the others; named layouts bit-identical): grid/named/lines 63.8µs → 54.1µs (−15%), grid/named/areas 66.6µs → 57.0µs (−14%), grid/named/unnamed 28.4µs → 26.0µs (−8%); items extending before the explicit grid via negative integer lines −6%, via negative named lines (-2 foo) −7%; grid/wide 31×31 / 100×100 / 316×316: −3% / −10% / −5%; grid/deep −4.5% / −8% / −5%; grid/superdeep −2% / +2%. The named benchmark (12×12 grid, 144 items placed by <integer> <ident> + span <ident>, by area name, or by negative lines) is not part of this PR.

Tests: new generated fixtures grid_implicit_tracks_named_line_placement (row flow) and grid_implicit_tracks_named_line_placement_auto_flow_column (column flow, so the estimate's role in the flow axis is covered for both axes); both fail on main.

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/2530fd358d4f4736bec3c4ac9accefdb
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/2530fd358d4f4736bec3c4ac9accefdb?variant=devin-insiders
Requested by: @nicoburns

[staging-devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'Devin'.

• Disable automatic comment, CI, and merge conflict monitoring