Phase 0 — Modularize the renderer

[ozgesolidkey]

Part of #14. Ref: docs/ARCHITECTURE_V2.md §6.

Carve src/renderer/renderer.ts (~7000 lines) into ES modules by feature, behavior-preserving. This is the gating step — features can't be extracted until the monolith has seams.

Tasks

• Add a renderer bundler (esbuild) so the renderer can use import/export while still loading as one renderer.js; keep tsc --noEmit for typechecking; main/preload stay CommonJS.
• Verify the bundled renderer is behavior-identical (build + 340 tests green + app smoke-test).
• Extract shared primitives into modules: state, elements, dom utils (escapeHtml, showToast), navigation (goToLine/jumpToTrendLine).
• Vertical-slice proof: extract the Investigate feature into its own module importing the shared primitives.
• Carve remaining features into modules incrementally (viewer, search, filter, analysis, trends, live, bookmarks, highlights, annotations, baselines, notes…).

Done when: renderer is a thin shell wiring feature modules; no single file > ~1500 lines; tests green.

Risk: low–medium (mechanical). Do NOT start L2/L3 extraction (other phases) before this lands.

[ozgesolidkey]

Foundation landed (branch refactor/v2-architecture): esbuild now bundles the renderer, wired into npm build, behavior-preserving (renderer still one script). Build + 340 tests green.

Key finding for the module conversion itself: adding a single import flips renderer.ts from a script to a module, which surfaces ~37 latent issues that must be handled deliberately (not a casual extraction):

• A flood of noUnusedLocals errors (TS6133) — TS doesn't enforce these on a global script, but does on a module. Several genuinely-dead functions (createLineElement, applyHighlights, performScrollUpdate, renderComparePanelDiff…) to remove or wire up.
• Inline interfaces stop merging with ambient ones in types.d.ts/shared. Today renderer.ts redefines SearchResult (with lineText), AnalysisResult (with insights), HighlightConfig/Bookmark (with isGlobal/lineText) and relies on declaration-merging with the global script. As a module these shadow/clash → type errors.

Implication: the conversion PR must (1) reconcile the duplicated interfaces into one shared source of truth, and (2) clean unused locals — before/as features are carved out. Do this with the app runnable for smoke-testing, since it touches type shapes the runtime relies on.