GitHub Agentic Workflows failing to consistently call dispatch workflow

The [Pipeline Analysis Next Steps](https://github.com/Azure/azure-sdk-for-python/blob/main/.github/workflows/pipeline-analysis-next-steps.md) workflow does not call the dispatch_workflow safe outputs step consistently. Instead, it calls the workflow name, which is turned into a MCP tool at runtime. As shown here: https://github.com/Azure/azure-sdk-for-python/actions/runs/33844761842/job/100935717375#process-safe-outputs

Log failure in safe output job:
```
Error: Failed to dispatch workflow "pipeline-analysis-auto-fix": Required input 'ci_head_sha' not provided - https://docs.github.com/rest/actions/workflows#create-a-workflow-dispatch-event
Error: ✗ Message 3 (dispatch_workflow) failed: Failed to dispatch workflow "pipeline-analysis-auto-fix": Required input 'ci_head_sha' not provided - https://docs.github.com/rest/actions/workflows#create-a-workflow-dispatch-event
```

Agent logs showcasing calling the wrong tool:
```
● dispatch_workflow (MCP: safeoutputs)
└ {"result":"Safe-job 'dispatch_workflow' executed successfully with arguments:...

● pipeline_analysis_auto_fix (MCP: safeoutputs) · ci_head_sha: "b33eecf66c6795359bd8d936bda40feb651…
└ {"result":"success"}
```

dispatch_workflow has empty required parameters:
```
{
"name": "safeoutputs-dispatch_workflow",
"namespacedName": "safeoutputs/dispatch_workflow",
"source": "mcp",
"title": "dispatch_workflow",
"description": "[safeoutputs] Custom safe-job: dispatch_workflow",
"input_schema": {
"additionalProperties": true,
"properties": {},
"type": "object"
},
"mcpServerName": "safeoutputs",
"mcpToolName": "dispatch_workflow",
"safeForTelemetry": {
"name": false,
"inputsNames": false
}
}
```

And at runtime, GH AW, creates a new safe output tool, instead of using the dispatch_workflow tool:
```
{
"name": "safeoutputs-pipeline_analysis_auto_fix",
"namespacedName": "safeoutputs/pipeline_analysis_auto_fix",
"source": "mcp",
"title": "pipeline_analysis_auto_fix",
"description": "[safeoutputs] Dispatch the 'pipeline-analysis-auto-fix' workflow with workflow_dispatch trigger. This workflow must support workflow_dispatch and be in .github/workflows/ directory in the same repository. Use the 'ref' parameter to target a specific branch or tag (allowed patterns: refs/heads/main).",
"input_schema": {
"additionalProperties": false,
"properties": {
"aw_context": {
"default": "",
"description": "Agent caller context (used internally by Agentic Workflows).",
"type": "string"
},
"ci_head_sha": {
"description": "Failed pull request commit",
"type": "string"
},
"parent_run_id": {
"description": "Trigger run that requested this fix",
"type": "string"
},
"pr_number": {
"description": "Pull request number to fix",
"type": "string"
},
"ref": {
"description": "The git ref (branch, tag, or SHA) to dispatch the workflow on. Must match one of the configured allowed ref patterns: refs/heads/main. If omitted, the dispatching workflow's ref is used.",
"type": "string"
}
},
"required": [
"ci_head_sha",
"parent_run_id",
"pr_number"
],
"type": "object"
},
"mcpServerName": "safeoutputs",
"mcpToolName": "pipeline_analysis_auto_fix",
"safeForTelemetry": {
"name": false,
"inputsNames": false
}
}
```

We should try moving Copilot requests to the Copilot SDK within the azsdk tool. This allows for more deterministic infrastructure to be built around the model's inputs and outputs.