Feature: In-memory background job system for Ollama installs with SSE streaming support

[DJJones66]

Feature: In-memory background job system for Ollama installs with SSE streaming support

---

Summary

This PR introduces a new in-memory background job system for handling Ollama model installations. The new system allows installs to run asynchronously, provides real-time feedback via server-sent events (SSE), and supports cancelation. It also improves resiliency in provider model lookups by falling back to direct SQL when ORM queries return no results.

---

Changes

🔹 AI Providers (ai_providers.py)

• Updated get_all_models:

  • Now prefers ORM-based get_all_parameterized.
  • Falls back to direct SQL get_all if ORM returns no settings.
  • Improves reliability across environments with legacy or mixed scope storage.

🔹 Ollama Endpoints (ollama.py)

• Added in-memory job model (InstallTask dataclass) with:

  • Task state tracking (queued, running, downloading, verifying, extracting, completed, error, canceled).
  • Progress, messages, error handling.
  • Subscriber support for event streaming with a ring buffer.
  • Deduplication logic to avoid duplicate installs for the same model/server.
  • Cleanup of old tasks after TTL expiry.

• Implemented background install worker:

  • Manages streaming downloads from Ollama /api/pull.
  • Emits structured events for state/progress.
  • Handles connection errors, server errors, cancelation, and success conditions.

• Added new API routes:

  • POST /install → enqueue background install, return task_id.
  • GET /install/{task_id} → retrieve current status.
  • GET /install/{task_id}/events → stream live events via SSE.
  • DELETE /install/{task_id} → cancel running task.

• Refined check_model_exists to use normalized server base URL.

• Normalized server base URL handling (normalize_server_base).

🔹 Frontend API Service (ApiService.ts)

• Added getSse helper:

  • Simple wrapper for GET-based SSE endpoints.
  • Automatically attaches bearer token if present.
  • Provides onMessage callback for streamed events.
  • Handles unexpected responses, connection errors, and closes cleanly.

---

Motivation

• The previous install endpoint tied the client request lifetime to long-running downloads, causing blocked UIs and unreliable progress reporting.

• The new system decouples installs from request lifetimes:

  • Users immediately get a task_id.
  • Install progress streams back via SSE or can be polled.
  • Tasks are tracked in-memory with state, progress, and error handling.

• Adds cancelation support for installs.

• Makes provider model fetching more robust by falling back to direct SQL queries when ORM scope filters fail.

---

Testing & Validation

• Verified new /install endpoint:

  • Returns task_id immediately.
  • Dedupes duplicate installs.
  • Streams progress events via SSE.
  • Polling fallback tested with /install/{task_id}.
  • Cancelation tested via DELETE /install/{task_id}.

• Confirmed tasks transition through correct states (queued → running → downloading → verifying → completed).

• Simulated server errors, timeouts, and malformed responses — errors surfaced in event stream.

• Frontend getSse tested against install events stream:

  • Correctly parses and forwards event data to callback.
  • Logs unexpected responses and errors for debugging.

---

Impact

• Backward-compatible: existing install endpoints replaced with async equivalents, but client integrations updated in ApiService to handle SSE.
• Greatly improves UX for model installs with responsive UI and real-time progress updates.
• Provides foundation for other background tasks to be managed in the same way.

---

Checklist

• Add in-memory job system for Ollama installs
• Implement background worker and event emission
• Add /install, /install/{task_id}, /install/{task_id}/events, /install/{task_id} DELETE endpoints
• Add cancelation and cleanup support
• Add frontend getSse helper for event streaming
• Improve ORM fallback for provider model lookups
• Validate across success, error, and cancelation scenarios