Add Arcee provider pricing

[pranav4501]

Summary

• add Arcee as a distinct inference provider
• add pricing and context windows for seven models served by Arcee
• resolve the provider from api.arcee.ai and explicit arcee provider names
• use exact model-ID matching to avoid collisions with the original model providers
• extract prompt, cached-prompt, and completion usage from Chat Completions responses
• regenerate the v2 catalog and bundled Python and JavaScript data
• add provider-resolution, pricing, and usage-extraction tests

Closes #569

Models

Model	Input / MTok	Cached input / MTok	Output / MTok
deepseek/deepseek-v4-flash-latest	$0.14	$0.028	$0.28
deepseek/deepseek-v4-pro	$1.74	$0.20	$3.48
deepseek/deepseek-v4-pro-0813	$1.32	$0.044	$3.96
moonshotai/kimi-k3	$3.00	$0.30	$15.00
thinkingmachines/inkling-small	$0.50	$0.10	$1.20
trinity-large-thinking	$0.25	$0.06	$0.80
zai-org/glm-5.2	$1.40	$0.26	$4.40

Sources

• Public pricing: https://docs.arcee.ai/get-started/pricing
• Models endpoint: https://api.arcee.ai/api/v1/models
• Chat Completions API: https://docs.arcee.ai/api-reference/chat-completion

The public pricing page lists six models. Pricing for deepseek/deepseek-v4-pro-0813 and the provider-specific context windows were verified through Arcee's authenticated /models endpoint.

Cache-write pricing and maximum output-token limits are omitted because Arcee does not currently publish them.

Validation

• make build
• make lint
• make typecheck
• Python: 842 passed, 41 expected failures, 1 deselected
• JavaScript: 1,852 passed, 1 skipped
• generated datasets are stable after regeneration
• git diff --check

The deselected Python test was the existing test_wait_prices_updated_async, which hangs indefinitely in this local environment. It is unrelated to the Arcee provider changes.

[cubic-dev-ai]

No issues found across 9 files

_{Re-trigger cubic}

[Kludex]

Superseded by #644, which follows the repository add-provider workflow. The replacement includes the public-source updater, recorded cassette, agentic monitoring, and Python/JavaScript parity tests.