Add message threading and per-message metadata view

[paullizer]

Summary:
Introduce a first-class threading model for conversations and a per-message metadata view in the UI. Threading will associate user messages with the hidden system prompt and assistant responses, tool/plugin invocations, and image generation events. The metadata view surfaces core context (agent, model, workspace, button states, IDs, timestamps, etc.) for each message. This enables edit/delete/retry flows with clear ripple-effect visibility while preserving prior generated content for cycling and audit.

Motivation:

• Today we only temporally connect messages by timestamps, which is fragile and doesn’t capture relationships or causality.
• Scientists/investigators need a holistic understanding of what was done, what came after, and how edits/ retries affect outcomes without losing prior outputs.
• Threading provides durable associations and ordering independent of timestamps and supports advanced actions like edit, delete, and retry.

Proposed Feature Set:

• Message Threading:

  • Associate user messages with the corresponding system message (hidden from chat but sent to the model) and the assistant response.
  • Link tool/plugin calls and image generation events to the originating message.
  • Connect message threads in series so we know sequence causality regardless of timestamps.
  • Track attempts per thread (e.g., retries increment attempt counters).
  • Preserve previous content while allowing cycling through versions.

• Per-Message Metadata View:

  • A structured metadata pane per message showing:
    • Conversation ID
    • Thread ID, Previous Thread ID, Active flag
    • Attempt number
    • Role (user/system/assistant/tool)
    • Timestamp
    • Agent details (display name, augmented flag)
    • Model details (Selected Model vs Frontend Model)
    • Workspace & Document Selection state
    • Button States (e.g., Image Generation: Disabled, Document Search: Disabled)
    • Global toggles (Search Enabled: Disabled, Global Agent: Disabled, Streaming: Disabled)
    • Chat Context summary
    • Optional: Avatar/profile info and message input context
  • A dedicated “Thread Information” block (Thread ID, Previous Thread, Active, Attempt).
  • A “Message Details” block (Conversation ID, Role, Timestamp).
  • A “Generation Details” block (Model, Agent).
  • Panels for “User Information”, “Agent Selection”, “Model Selection”, and “Workspace & Document Selection”.
  • Consistent card layout for agent outputs (e.g., an “Agent output card” with summary fields).

UI/UX Design Notes:

• Top-level header reflects the current operation (e.g., “find candidate with id …”) — keep generic to any data domain.
• Left/center: a large info panel grouping:
  • User Information
  • Thread Information
  • Button States
  • Workspace & Document Selection
  • Agent Selection
  • Model Selection
  • Chat Context
• Main content: agent output card that presents a structured summary.
• Bottom/right: small avatar, thread details block, and message input area/bottom toolbar.
• Toggle visibility for Streaming, Image Generation, Document Search states in Button States panel.
• Distinguish Selected Model from Frontend Model and show Agent Augmented flag.

Data Model Updates:

• Thread entity:
  • id
  • previous_thread_id (nullable)
  • active (bool)
  • attempt (int)
  • created_at, updated_at
• Message entity additions:
  • thread_id (FK)
  • role (user/system/assistant/tool)
  • series_index (integer ordering within thread; independent of timestamp)
  • linked_message_ids (associations to system/assistant/tool/image-gen messages)
  • metadata (JSON: agent, model, workspace, button states, toggles, etc.)
  • deleted_at (nullable for soft delete)
• Ordering rules:
  • Primary order by series_index within thread; timestamps are secondary.

Behavior:

• Mask:
  • Help provide more control over what the model receives as part of conversation history. You may like some parts of the models response but not want all of it.
  • Mask selected parts of a message to remove them from being sent to the model on subsequent submissions.
  • Mast entire message to remove it from being sent to the model on subsequent submissions.
• Edit:
  • Creates a new message version in the same thread; series_index advances.
  • Prior content remains accessible; UI shows ripple warning indicating downstream messages may diverge.
• Delete:
  • Soft-deletes a message; thread remains intact; downstream associations remain but flagged.
• Retry:
  • Increments thread attempt; generates new assistant/tool outputs linked to the originating user message.
• Image/tool linkage:
  • Tool and image-generation messages are explicitly linked to their triggering message and appear in the thread chain.