feat(eval): add read-only Eval TUI

[jariy17]

What

Adds the interactive TUI over the existing eval command tree, scoped to read-only screens. It opens a working menu → list → detail → JSON flow, matching runtime/memory.

Screens

Path	Screen
eval, eval evaluator, eval online-eval	RouterScreen menus
eval evaluator list / online-eval list	PaginatedTablePicker
eval evaluator get [/json]	ResourceDetailScreen hub → JsonDetail
eval online-eval get [/json]	ResourceDetailScreen hub → JsonDetail

Notes for reviewers

• No Core changes, no new AWS calls, no new deps. Every screen reuses landed components (ResourceDetailScreen from refactor(tui): share resource detail screen #1890, PaginatedTablePicker, JsonDetail, RouterScreen). Only genuinely new code is two pickers + four screen files + three thin menu screens.
• RouterScreen gains an omit?: string[] prop. The eval command tree still contains the mutation commands, and RouterScreen builds its menu from that tree. Without filtering, a mutation item in the menu would route to nothing and fall through to the * → HelpScreen catch-all, which calls exit() and tears down the app. omit hides them from the menu; they remain usable from the CLI. Backwards-compatible (defaults undefined; existing menus unchanged). Read-only, llm-as-a-judge/code-based have only create/update children, so those submenus are omitted too — evaluator shows just get/list.
• Evaluator kind (LLM-as-a-Judge vs code-based) is derived from which arm of the evaluatorConfig union is populated — the GetEvaluator response has no type field of its own (unlike the list summary's evaluatorType).
• Online-eval sampling reads rule.samplingConfig.samplingPercentage; sampling/evaluator-count are only on the get response, not the list summary, so they live on the detail screen.

Behaviour change

Mounting withTuiOnEmptyFlagsAndArgs means a bare read command with no flags (e.g. eval evaluator get) now opens the TUI instead of erroring required option '--id' not specified. This matches runtime/memory. Existing leaf tests were updated to assert the headless path under --json (same convention as the memory read-only TUI), plus new tests assert bare read leaves open the TUI.

Testing

• New: evaluator.screen.test.tsx, online-eval.screen.test.tsx — menus omit mutations, lists render/paginate/select, detail hubs render + route to JSON, retry, empty states.
• Updated eval leaf tests for the --json headless contract.
• Full suite green (720 tests), typecheck + lint clean.
• Verified live against real us-west-2 data through the tui-harness (menu → list → detail hub → JSON, both resources).

Plan docs: read-only spec and Phase 2 (mutations) spec are tracked separately.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.28%. Comparing base (206540c) to head (4c0712c).

Additional details and impacted files

@@             Coverage Diff              @@
##           refactor    #1920      +/-   ##
============================================
+ Coverage     96.17%   96.28%   +0.10%     
============================================
  Files           236      245       +9     
  Lines         11778    12105     +327     
============================================
+ Hits          11328    11655     +327     
  Misses          450      450              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.