fix(azure): preserve model-to-deployment mapping for images.edit

[serhiizghama]

Problem

Fixes #1917

On Azure, images.edit() ignores the per-request model param for deployment routing. With no client-level deployment, the request goes to /openai/images/edits instead of /openai/deployments/{model}/images/edits and Azure returns 404. Every other deployments-routed endpoint (chat completions, embeddings, images.generate, audio transcriptions/translations) handles this correctly.

Root cause: AzureOpenAI.buildRequest resolves the deployment via

const model = this.deploymentName || options.body['model'] || options.__metadata?.['model'];

/images/edits is multipart, so options.body is already a FormData when buildRequest runs and options.body['model'] is undefined. The multipart audio endpoints solve exactly this by passing __metadata: { model: body.model } — images.edit is the only entry in _deployments_endpoints with a multipart body that doesn't.

Solution

Pass __metadata: { model: body.model } in images.edit, mirroring the existing pattern in audio/transcriptions.ts and audio/translations.ts. No behavior change for non-Azure clients (__metadata is only read by AzureOpenAI.buildRequest).

Since this file is generated, the underlying fix may need to land in the Stainless config — the change here reproduces exactly what the generator already emits for the audio multipart endpoints, so it should be a straightforward config mirror.

Testing

Added handles image edit to the existing "with no client-level deployment" suite in tests/lib/azure.test.ts (the suite covered images.generate but not images.edit, which is how this slipped through). The test fails on master with:

- "url": "https://example.com/openai/deployments/deployment/images/edits?api-version=..."
+ "url": "https://example.com/openai/images/edits?api-version=..."

and passes with the fix. Full offline suite (jest --testPathIgnorePatterns tests/api-resources): 567/567 passing; lint, prettier and ./scripts/build clean.