fix: omit unsupported GPT-5 chat params for Azure Foundry custom providers

[itzAmrit]

Refs #27948

This strips the GPT-5 chat parameters that Azure Foundry rejects when a custom provider uses @ai-sdk/openai-compatible against an Azure host.

What changed:

• omit max_tokens by clearing maxOutputTokens
• omit reasoningSummary for the same Azure OpenAI-compatible GPT-5 path
• add a plugin regression test for Azure hosts

Verification:

• bun --cwd packages/opencode test test/plugin/azure.test.ts test/plugin/cloudflare.test.ts
• repo pre-push hook passed bun turbo typecheck

Tested on:

• macOS 26.4.1 (build 25E253)
• Apple Silicon / arm64
• bun 1.3.14
• node v26.0.0
• local OpenCode repro was on opencode 1.14.48 against Azure Foundry GPT-5 chat deployments

[itzAmrit]

Closing this because the local setup was corrected to use the built-in azure provider with resourceName, rather than the custom @ai-sdk/openai-compatible Azure Foundry route. With the correct Azure config (and without forcing a full /openai/v1 base URL), both azure/gpt-5.4 and azure/gpt-5.3-codex work locally, so this patch is no longer the right fix.