feat(integration): add allow_all_models attribute to control model access

[austin-engle]

Summary

Adds an allow_all_models boolean attribute to the portkey_integration resource, allowing users to control whether all models are enabled by default for an integration.

• Problem: When an integration is created, the Portkey API defaults allow_all_models = true, meaning every model for that provider is available. There was no way to restrict model access to only explicitly enabled models via portkey_integration_model_access resources.
• Solution: Expose the existing allow_all_models API field on the portkey_integration resource. Setting it to false makes portkey_integration_model_access resources act as an allowlist.

Usage

resource "portkey_integration" "openai" {
  name             = "OpenAI"
  ai_provider_id   = "openai"
  allow_all_models = false  # Only explicitly enabled models are available
}

resource "portkey_integration_model_access" "gpt5_mini" {
  integration_id = portkey_integration.openai.slug
  model_slug     = "gpt-5-mini"
  enabled        = true
}

Implementation Details

• The allow_all_models field is managed via the models endpoint (PUT /integrations/{slug}/models), not the integration CRUD endpoint. The provider makes a separate API call when needed.
• Defaults to true to match existing API behavior (non-breaking change).
• On Create: only calls the models endpoint when allow_all_models = false (skips redundant call for default).
• On Create: sets integration state before the models API call so partial failures don't leave orphaned resources.
• On Read: fetches current value from GetIntegrationModels() for drift detection.
• On Update: only calls models endpoint when the value actually changed.

Changes

• internal/provider/integration_resource.go — Added schema attribute, Create/Read/Update logic
• internal/provider/integration_resource_test.go — Added acceptance tests for create-with-false, toggle true↔false, default behavior assertion
• docs/resources/integration.md — Updated with new attribute docs and example
• CHANGELOG.md — Added unreleased entry

Test plan

• go fmt, go vet, go build all pass
• Existing acceptance tests unaffected
• New acceptance tests: create with allow_all_models = false, toggle true→false→true
• Default behavior: existing configs without allow_all_models get true (non-breaking)
• Import roundtrips correctly (Read fetches value from API)

[roh26it]

Approved ✓

Reviewed all code changes and ran acceptance tests locally. Everything looks good.

Review Summary

Code Quality:

• Build, vet, and format all pass
• Schema follows existing patterns (Optional/Computed/Default)
• Error handling is consistent with other resources
• Efficient implementation - only calls models API when needed

Test Coverage:

• TestAccIntegrationResource_allowAllModelsFalse - PASS
• TestAccIntegrationResource_allowAllModelsUpdate - PASS
• TestAccIntegrationResource_basic - PASS (includes import verification)
• Tests check actual values, not just existence

Implementation Details I Appreciated:

• Partial failure handling: state is set before the models API call so Terraform tracks the resource even if that call fails
• Non-breaking default (true) matches existing API behavior
• Reads from API on each operation for proper drift detection

Nice work on the PR description too - clear problem statement, implementation details, and test plan.