[Hackathon] feat: add Texera Assistant floating panel (notifications, AI, search)

[jaeyun0503]

What changes were proposed in this PR?

Adds a Texera Assistant — a floating, draggable, resizable panel that lives in every page (logged-in users only) and surfaces workflow, social, and admin activity, plus first-class AI agent integrations and global search.

Demo Video

https://drive.google.com/file/d/1HeDvcvLqs_d2nEf8uhtVr1Z9Oo0tokKH/view?usp=sharing

Sample Screenshots

New UI surface

• Floating launcher (robot icon) — draggable around the viewport; the panel adapts which corner it opens into based on available space and is user-resizable from any free edge.
• Three tabs for regular users (Notifications, Workflows) plus a fourth Requests tab for admins. Each tab has its own Clear button and the Clear state persists across reloads.
• Settings overlay (gear icon) — per-type toggles to mute specific notification kinds (successful runs, dataset likes, etc.).
• Search overlay (search icon) — calls the existing dashboard search API and lets users jump to any workflow or dataset they can access, with creation/modified dates inline.

Notification feeds

• Run notifications — push on Completed / Failed / Killed transitions, with workflow-aware name capture (survives navigation), Retry / Kill actions, and a dedup window to handle websocket replays after HMR or refresh.
• Social notifications — polls the hub for likes / clones on the user's workflows and datasets; baseline-on-first-poll so users don't get spammed on login.
• Admin requests — polls for INACTIVE users and uses a new DB-backed request_viewed flag (see schema changes below) so the "viewed" state is global across admins and survives reloads. Clicking a request marks it viewed and navigates to the admin user page; Clear marks every currently-pending request viewed in one transaction.

AI agent integration

• Per-workflow agent binding — AgentPanelComponent now persists which agent the user last activated, keyed by workflow id, so re-opening a workflow restores the right agent (or shows the registration tab if none is bound). Switching workflows correctly swaps the active agent.
• AI failure explanations — when a workflow fails, the assistant looks up the bound agent for that workflow, sends a structured prompt with operator-level error details, and streams the agent's reply into the notification as rendered markdown. The reply also appears in the AI panel chat. Conversational follow-ups (questions, offers to "fix it for you") are trimmed via prompt instructions plus a post-processor so the notification stays focused on causes and remediation steps.
• Operator tab (workflow page only) — shows the currently selected operator's type and properties with an Explain with AI button that streams a markdown explanation. Explanations are cached per operator id, and the tab is "sticky" — deselecting an operator on the canvas doesn't wipe the view.
• Coordinated launcher buttons — a new AgentPanelControlService keeps the floating "AI Agent Builder" flask button in sync with the existing agent panel's open state; clicking either dismisses the other.

Persistence

Several pieces of state are persisted to localStorage so reloads don't reset the user's view:

• Notifications, dismissed terminal-state dedup entries, session workflow list (with per-wid dismissal), drag position, panel size, notification settings, and the workflow → agent map.

Schema / backend changes

• Adds request_viewed BOOLEAN NOT NULL DEFAULT FALSE to the user table (sql/texera_ddl.sql + sql/updates/23.sql). The migration backfills all existing rows to TRUE so admins don't get flooded with notifications for accounts that pre-date this feature.
• Updates the JOOQ-generated User table descriptor with the new column.
• AdminUserResource exposes the field in /admin/user/list and adds two endpoints:
  • POST /admin/user/mark-requests-viewed (per-uid)
  • POST /admin/user/mark-all-requests-viewed (bulk)
• Frontend User type, AdminUserService, and the assistant all consume the new field.

How was this PR tested?

Manually Tested.

Was this PR authored or co-authored using generative AI tooling?

Generated-by: Claude 4.7

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 42.88%. Comparing base (7bb102e) to head (53e6d0a).
⚠️ Report is 9 commits behind head on main.

Additional details and impacted files

@@             Coverage Diff              @@
##               main    #5093      +/-   ##
============================================
+ Coverage     42.85%   42.88%   +0.03%     
- Complexity     2207     2209       +2     
============================================
  Files          1045     1045              
  Lines         40146    40141       -5     
  Branches       4240     4240              
============================================
+ Hits          17203    17214      +11     
+ Misses        21878    21860      -18     
- Partials       1065     1067       +2     

Flag	Coverage Δ		*Carryforward flag
access-control-service	39.53% <ø> (ø)		Carriedforward from 7bb102e
agent-service	33.72% <ø> (ø)		Carriedforward from 7bb102e
amber	43.74% <ø> (+<0.01%)	⬆️	Carriedforward from 7bb102e
computing-unit-managing-service	0.00% <ø> (ø)
config-service	0.00% <ø> (ø)
file-service	32.18% <ø> (ø)
frontend	33.93% <ø> (+<0.01%)	⬆️	Carriedforward from 7bb102e
python	88.99% <ø> (ø)		Carriedforward from 7bb102e
workflow-compiling-service	56.81% <ø> (+9.09%)	⬆️

*This pull request uses carry forward flags. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 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.

[mengw15]

Please modify the title to follow the Hackathon title convention