Update W&B active model catalog

[alexeldeib]

Summary

Updates the Weights & Biases provider catalog to match the current authenticated /v1/models active model list.

This changes the W&B generator so it uses:

• authenticated https://api.inference.wandb.ai/v1/models as the source of truth for enabled model IDs
• W&B Artificial Analysis metadata as the rich metadata source for model TOMLs
• explicit manual/probed overrides for active models missing from the AA metadata feed

OpenRouter metadata is intentionally not added because it currently provides no unique active IDs or differing metadata for W&B.

Changes

• Adds 9 active W&B models:
  • Qwen/Qwen3.5-27B
  • Qwen/Qwen3.5-35B-A3B
  • Qwen/Qwen3.6-27B
  • Qwen/Qwen3.6-35B-A3B
  • deepseek-ai/DeepSeek-V4-Flash
  • deepseek-ai/DeepSeek-V4-Pro
  • google/gemma-4-31B-it
  • ibm-granite/granite-4.1-8b
  • moonshotai/Kimi-K2.6
• Removes inactive zai-org/GLM-5-FP8
• Refreshes existing W&B model metadata, including cache-read pricing and last_updated
• Corrects W&B gpt-oss models to open_weights = true
• Adds code comments documenting live /v1/chat/completions probes used for support that is not exposed by AA metadata
• Makes the generator fail if a future active W&B model lacks both AA metadata and an explicit override

Verification

WANDB_API_KEY=... WANDB_INFERENCE_PROJECT=wandb/inference-test bun packages/core/script/generate-wandb.ts --dry-run
# 0 would be created, 0 would be updated, 0 would be removed, 26 unchanged

bun validate

cd packages/web && bun run build

git diff --check