feat(language-server): standalone Malloy LSP with render-tag diagnostics

[jswir]

Summary

• Add @malloydata/malloy-language-server — a standalone LSP-over-stdio implementation for .malloy, .malloysql, and .malloynb files. Provides diagnostics, document symbols, code lenses, completions, hover, go-to-definition, and code actions.
• Ship a CLI binary (malloy-language-server --stdio) and a Claude Code plugin scaffold under claude-code-plugin/ so LSP-capable editors (Neovim, Emacs, Helix, Claude Code, etc.) can spawn the server directly.
• Wire @malloydata/render-validator into the diagnostic pipeline so renderer tag mistakes (e.g. # currency on a string field) surface in the editor as the user types instead of only at preview time. Render-validator runs the renderer's own validation against the compiled result without executing the query or needing a DOM. Diagnostics carry code: render-tag so they're distinguishable from compile errors.

Test plan

• npm run build builds the new package cleanly against current main
• npx eslint 'packages/malloy-language-server/src/**/*.ts' passes (0 errors; pre-existing no-console warnings only)
• All new files carry the standard copyright header
• Stdio smoke test: initialize → initialized → workspace/didChangeConfiguration → textDocument/didOpen round-trips with capabilities and emits publishDiagnostics
• Render-tag diagnostic verified end-to-end with the smoke test:

  query: q is duckdb.sql("SELECT 'hello' as val") -> {
    select:
      # currency
      val
  }
  

  → [sev=Error] code=render-tag Tag 'currency' on field 'val' requires a numeric field, but field is string
• Reviewer: exercise the server in an LSP client (Neovim, Helix, VS Code generic-LSP, Claude Code)