feat: inject _instance_name from provider name field into config

[ekg]

Summary

• amplifier_core/session.py: Inject _instance_name from provider name field into config before loading the provider

When the provider config has a 'name' field (from settings.yaml), inject it as '_instance_name' into the provider config before loading. This enables named provider instances where the same module (e.g., provider-openai) can be used for multiple services (openrouter, deepseek, etc.) with different names.

Context

Companion to PR #17 on provider-openai which added _instance_name support for named provider instances.

Test plan

• Provider with name field gets _instance_name injected
• Provider without name field falls back to module ID
• Multiple providers with same module but different names work correctly

🤖 Generated with Amplifier

[bkrabach]

Thanks for the idea, @ekg. Named multi-instance provider support has since shipped in v1.0.12 via PR #41 and #42 — providers now get proper instance_id extraction, multi-instance validation, and snapshot-based remapping. The _instance_name injection approach in this draft was heading in the right direction, but the final implementation took a more robust path. Closing as superseded by the instance_id architecture. If you need a separate human-readable display name on top of instance_id, happy to discuss a new PR against the current codebase.