feat: features list component, text highlighting, and multi-table support

[donghaoren]

Features List chart

• New Features List chart: searchable, sortable, filterable panel of features with per-feature counts, bars, and selection.
• Binary and multiclass prediction bars showing how predictive each feature is of a target column.
• Source-aware counts, source-skew ranking, feature pinning, and optional grouping by topics.

Highlight layer

• Highlight relevant part of the texts in the instances view when a feature is selected. Example below: highlighting text related to "pet care"
• Rewrite the code for text highlighting to allow using embedding similarity for highlighting (in addition to the existing fuzzy text match).

Embedding layer

• Multi-provider client-side embedding: transformers.js (local) and OpenAI-compatible servers (users need to set this up themselves).
• Refactored the code to achieve lower peak memory on large datasets; projection orchestration split into its own module.

Multi-table support

• Backend: new --table NAME PATH CLI option (repeatable) to load auxiliary tables alongside the main dataset.
• Frontend: additionalTables prop with join config for cross-table filtering and highlighting.
• Charts can render from non-primary tables.

Settings & UI

• New provider-config UI: endpoint, API key, and model selection, persisted locally in the browser.