fix(#179): move data-skipping indexes from schema graph cards to the detail drawer

[BorisTyshkevich]

What & why

Closes #179. Follow-up to the #177 compact-type work.

Expanded schema-graph cards used to append an inline idx: … line. Since a
card's width is its widest rendered text line, long index names/types inflated
the card and distorted graph layout — a heavily-indexed table spread far wider
than the same columns with no indexes, pushing related nodes apart. The graph
card is an overview surface; a flattened, capped index list is detailed metadata
that doesn't belong in card geometry.

This removes indexes from the card entirely (card width/height/layout/spacing are
now independent of index count, name, and type) and shows the complete index
metadata in a new drawer section.

Schema graph card

• No more idx: line; no .eg-skipidx element or CSS.
• Card model (schema-cards.js) drops MAX_IDX / idxOverflow / skipLine;
  buildCardModel/buildCardGraph no longer take or pass skip indices, and
  cardSize() no longer counts an index row.

Bottom detail drawer (schema-detail.js)

• New Data-skipping indexes (N) section after Columns, before Partitions.
• Shows every index (never a capped subset): Name, Expression, full Type,
  Granularity, and compressed size.
• type_full is preferred over type so parameterised indexes
  (bloom_filter(0.01), tokenbf_v1(30720, 4, 0)) stay distinguishable; falls
  back to type when absent.
• Long Name/Expression/Type cells ellipsize with the full value on hover (the
  same cappedCell idiom the column cells already use — no new CSS).
• Section omitted entirely when the table has no skipping indexes; missing
  fields (granularity/sizes) degrade safely.

Data source (ch-client.js)

• loadTableDetail now returns { columns, indexes, partitions, ddl, comment },
  fetching the index rows once per detail-open in its existing parallel batch
  (name, expr, type, type_full, granularity, compressed/uncompressed/marks from
  system.data_skipping_indices).
• Not reused from the schema-graph payload: that payload carries only
  name/type/expr, can't reach the drawer's click handler without threading
  arrays through the dagre carry() layout (the "worse coupling" the issue
  warns against), and the drawer requires type_full + granularity. The
  issue's escape hatch ("extending the existing detail query is acceptable when
  reuse would create worse coupling") applies.
• The now-dead skip-index read loadSchemaCards issued on every graph load
  is removed (one fewer query on graph open).

Checklist

• npm test passes (per-file coverage gate; schema-detail.js 100/100/100/100)
• Tests added/updated in the same change (card model, graph render, drawer, network)
• npm run build succeeds (single-file dist/sql.html)
• Layers kept honest: pure model in src/core/, fetch in src/net/, DOM in src/ui/
• No new runtime dependency
• README + CHANGELOG.md [Unreleased] updated (grants table + Changed entry)
• Reconciled: Move data-skipping indexes out of schema graph cards and into the detail drawer #179 is a standalone bug, not a roadmap-Roadmap to 1.0.0 #68 track item

Verification

Unit tests assert the exact rendered DOM (section ordering, headers, all rows,
type_full preference + fallback, ellipsis + title, empty-table omission).
Additionally rendered in a real browser via a throwaway harness with sample OTel
index data to confirm layout/alignment/ellipsis — screenshot matched.

🤖 Generated with Claude Code