feat: add OpenShift deployment and numeric reasoning controls

[gprocunier]

Summary

Adds a production-oriented OpenShift deployment path for Odysseus and endpoint-scoped continuous reasoning controls for compatible local models.

• restricted-SCC-compatible Odysseus, ChromaDB, and SearXNG deployment packaging
• Kustomize base and example overlays, rollout documentation, readiness handling, and scoped private-CA support
• endpoint metadata for models that accept a continuous reasoning-effort value
• a 0.00 to 0.99 thinking-effort slider for endpoints with the float capability enabled
• numeric effort passed through chat_template_kwargs on synchronous, asynchronous, streaming, and tool-call requests
• unchanged behavior for endpoints that use named reasoning levels or do not expose reasoning controls

Why

Some model chat templates accept a continuous reasoning-effort value while OpenAI-compatible server schemas restrict the top-level reasoning_effort field to named levels. Passing an explicitly configured float through chat_template_kwargs lets the model template receive the intended value without model-name heuristics or changes to other providers.

The control is capability driven and endpoint scoped. It is rendered only when reasoning_effort_type is float.

Linked Issue

OpenShift deployment work is part of #5037.

Target Branch

• This PR targets dev.

Type of Change

• Bug fix
• New feature
• Breaking change
• Refactor / cleanup
• Documentation
• CI / tooling / configuration

Checklist

• I searched open issues and pull requests and found no duplicate continuous-effort implementation.
• This PR targets dev.
• The commits are limited to the OpenShift deployment and endpoint reasoning-control scope described above.
• Both paths were exercised in running OpenShift deployments in addition to the automated checks below.

How to Test

1. Render the OpenShift base and example overlay:

   kubectl kustomize deploy/openshift/base
   kubectl kustomize deploy/openshift/overlays/example
   oc apply --dry-run=client -k deploy/openshift/overlays/example -o name

2. Configure a compatible endpoint with a continuous effort value:

   PATCH /api/model-endpoints/{endpoint_id}
   Content-Type: application/json
   
   {"reasoning_effort_type":"float","reasoning_effort":0.9}

3. Confirm Admin > Added Models displays the slider and that chat and tool-call requests contain:

   {"chat_template_kwargs":{"reasoning_effort":0.9}}

4. Run the focused checks:

   python -m pytest -q tests/test_model_routes.py tests/test_llm_core_*.py
   node --check static/js/admin.js

Validation

• 326 model-route and LLM-core tests passed, including 7 focused continuous-effort cases
• Python syntax compilation passed for the changed backend and test files
• node --check static/js/admin.js passed
• the continuous-effort path was exercised against a vLLM-served model through a multi-round tool call

Visual Change

Admin > Added Models gains a compact thinking-effort slider only for endpoints configured with the float capability. Other endpoint rows are unchanged.

[gprocunier]

@pewdiepie-archdaemon from my testing, effort for thinkingmachine models like inkling will need the float slider if you want to use them properly.

https://huggingface.co/thinkingmachines/Inkling