Support Azure OpenAI deployment-name resolution in the router

[navintkr]

Azure OpenAI doesn't surface the underlying model id through
vscode.lm — it surfaces a deployment name chosen by whoever
provisioned the resource (e.g. gpt-4o-prod, my-team-mini). Right
now Token Proctor's router only matches on canonical ids, so Azure
users see their deployments as "unknown" and get routed to whatever
non-Azure model is available.

What to do

1. Add an optional azureDeployments section to .token-proctor.json
   that maps deployment name → canonical model id, e.g.:

   {
     "azureDeployments": {
       "gpt-4o-prod": "gpt-4o",
       "my-team-mini": "gpt-4o-mini"
     }
   }

2. Update src/core/policy.ts to parse the new
   field.
3. In src/core/modelRouter.ts, when
   resolving a model id that isn't in the catalog, check the deployment
   map before marking it unknown.
4. Document it in the README policy block.

Acceptance

• New config field parsed without breaking existing policies.
• A deployment name like my-team-mini gets costed as
  gpt-4o-mini in the estimate.
• Unit test covering a policy with two deployments.

Why this matters

Teams on Azure-hosted Copilot / Foundry are exactly the audience that
cares most about cost visibility — and today they get the worst
experience.