[All Platforms][Inference] inference set --provider ollama_local (underscore) rejected instead of normalized to ollama-local

## Description

`nemohermes {sandbox} inference set --provider ollama_local ...` (installer-style underscore spelling) is rejected outright with "Unsupported provider" instead of being normalized to the canonical hyphenated `ollama-local`, contradicting the documented/expected normalization behavior for this exact installer-style input pattern.

**Platform scope:** Reproduced on DGX Spark (aarch64) only; other platforms not tested this pass.
**Regression:** Unknown — earlier versions not tested this pass.

## Environment

- Device: DGX Spark (real GB10 GPU)
- OS: Linux aarch64
- Node.js: v22.22.1
- npm: 10.9.4
- Docker: 28.3.3
- OpenShell CLI: 0.0.106
- NemoClaw: v0.0.121

## Steps to Reproduce

1. Onboard a Ready NemoHermes sandbox with local Ollama (any model), e.g.: `nemoclaw onboard --name hermes-dash --agent hermes --no-sandbox-gpu --yes --yes-i-accept-third-party-software`
2. Run: `nemohermes hermes-dash inference set --provider ollama_local --model qwen2.5:1.5b`

## Expected Result

The underscore-spelled provider input is accepted and normalized to the canonical `ollama-local`; the command exits 0 and both Hermes config surfaces use the normalized provider and the requested model.

## Actual Result

```
Exit code 2.
Unsupported provider 'ollama_local'. Supported providers: nvidia-prod, nvidia-nim, nvidia-router, openai-api, openrouter-api, anthropic-prod, compatible-anthropic-endpoint, gemini-api, compatible-endpoint, hermes-provider, ollama-local, vllm-local.
```

The command rejects the underscore form outright rather than normalizing it, even though `ollama-local` (hyphenated) is listed as a valid provider in the same error message.