[src] Hardcoded / legacy Bedrock model IDs in src/agentic_platform/ should be centralised

[ManojRamani]

Problem

src/agentic_platform/ has Claude and Nova model IDs hardcoded inline across several files. At least one of these (BasePrompt.model_id) is a legacy Bedrock model ID — calls to it now fail with:

```
ResourceNotFoundException: Access denied. This Model is marked by provider as Legacy and you have not been actively using the model in the last 30 days. Please upgrade to an active model on Amazon Bedrock
```

Even the non-legacy IDs are drift hazards: the same bump chore we just tackled in labs/ (tracked in #50) will recur here whenever a model hits end-of-life.

Occurrences

Legacy / known-problematic

• `src/agentic_platform/core/models/prompt_models.py:4` — `HAIKU_MODEL_ID = "us.anthropic.claude-3-haiku-20240307-v1:0"` (legacy; this is the one that triggered the ResourceNotFoundException from a lab notebook).
• `src/agentic_platform/agent/agentic_chat/prompt/agentic_chat_prompt.py:60` — same legacy Haiku 3 ID.

Other hardcoded IDs (not necessarily legacy today, but still drift-prone)

• `src/agentic_platform/tool/retrieval/retrieval_tool_prompt.py:21` — `us.amazon.nova-lite-v1:0`
• `src/agentic_platform/agent/agentic_rag/prompt/agentic_rag_prompt.py:70` — `anthropic.claude-sonnet-4-20250514-v1:0`
• `src/agentic_platform/agent/agentic_rag/agent/agentic_rag_agent.py:30` — `anthropic.claude-sonnet-4-20250514-v1:0`
• `src/agentic_platform/agent/agentic_chat/agent/agentic_chat_agent.py:34` — `anthropic.claude-sonnet-4-20250514-v1:0`
• `src/agentic_platform/agent/jira_agent/jira_agent.py:29` — `us.anthropic.claude-sonnet-4-20250514-v1:0`
• `src/agentic_platform/agent/jira_agent/jira_prompt.py:14` — `anthropic.claude-sonnet-4-20250514-v1:0`
• `src/agentic_platform/agent/strands_glue_athena/agent_service.py:49` — `us.anthropic.claude-3-7-sonnet-20250219-v1:0`
• `src/agentic_platform/service/litellm_gateway/litellm_config.yaml` — multiple model-name / routing entries, several tied to older Sonnet 3.5 / Haiku 3 IDs.

Impact

• `BasePrompt` inherits a legacy Haiku 3 default, so every downstream class that doesn't override `model_id` fails at runtime. This surfaced when a lab notebook (`labs/module2/notebooks/3_routing.ipynb`) called `bedrock.converse()` via `prompt.model_id`.
• Each time Bedrock deprecates a model, a maintainer has to grep `src/` and bump N files one by one — same pattern as [labs] Notebooks break recurrently due to hardcoded legacy Bedrock model IDs #50 was raised for on the lab side.
• `litellm_config.yaml` drifts from the Python defaults silently (different teams edit it at different times).

Proposed fix

Mirror the approach that's now in place for `labs/`:

1. Single source of truth in `src/` — add `src/agentic_platform/core/models/model_config.py` (or promote the existing `HAIKU_MODEL_ID` constant into a real config module) with constants for each model role:

   ```python
   HAIKU_MODEL_ID = "us.anthropic.claude-haiku-4-5-20251001-v1:0"
   SONNET_MODEL_ID = "us.anthropic.claude-sonnet-4-5-20250929-v1:0"
   NOVA_LITE_MODEL_ID = "us.amazon.nova-lite-v1:0"
   ```

2. Replace hardcoded literals in every `prompt_models.py` / `_agent.py` / `_prompt.py` with imports from that module.

3. Keep `litellm_config.yaml` in lockstep — either template it from the Python constants at build time, or add a CI check that fails if the two disagree.

4. (Optional) drift check — extend the existing `scripts/migrate_model_ids.py` audit to also scan `src/` for hardcoded `anthropic.claude-` / `amazon.nova-` literals. That script is used by the labs cleanup and already does the right thing for notebooks; adding `.py` / `.yaml` coverage would make it CI-ready.

Context

• Scope for this issue is deliberately limited to `src/agentic_platform/` — the lab-side cleanup is complete (lab notebooks now import from `labs_common`, and a `LabsBasePrompt` subclass overrides the legacy default from the current `BasePrompt`).
• The lab-side fix is intentionally read-only from `src/`'s perspective: it subclasses `BasePrompt` and pins `model_id`, rather than mutating the platform class. This issue is for the properly-scoped follow-up inside `src/` itself.
• Related issues: [pyproject] Clean up stale uv.workspace member paths #48 (stale `tool.uv.workspace` paths), [labs] Notebooks break recurrently due to hardcoded legacy Bedrock model IDs #50 (notebook-side cleanup).