[Backlog] Document new AI pricing workloads

[CheeYuTan]

Summary

Add user-guide and reference documentation for recently implemented AI pricing workloads that are available in Lakemeter but are not yet covered by the docs site.

This follows the documentation gap identified by @Menotron in #27.

Scope

Document the following workloads:

• Unity AI Gateway ([Backlog] Add AI Gateway pricing workload #26)
• Agent Evaluation ([Backlog] Add Agent Evaluation pricing workload #27)
• AI Runtime / Model Training (Add AI Runtime model training pricing #32)

For each workload, cover:

• Form inputs, units, defaults, and supported input methods
• Monthly usage and DBU calculation formulas
• Billing SKU and exact regional price behavior
• Important assumptions and excluded/additive costs
• Excel export representation
• Related workloads that users may need to estimate separately

Important workload-specific notes:

• AI Gateway: document Inference Tables and Usage Tracking separately, including their independent payload inputs.
• Agent Evaluation: explain that evaluation service charges are additive and exclude evaluated application or model-inference costs.
• AI Runtime: document A10/H100 accelerator choices, GPU-hour conversion, the MODEL_TRAINING SKU, price-driven AWS/Azure availability, and why GCP is not currently offered in the form.

Acceptance criteria

• Add one user-guide page per workload, following the structure of the existing AI Parse and AI Extract pages.
• Add all pages to the documentation sidebar and workload overview.
• Update the calculation reference and API reference where applicable.
• Include worked examples that reconcile monthly usage, DBUs, SKU rate, and total cost.
• Link assumptions to official Databricks pricing or product documentation.
• Verify terminology and formulas against the implemented calculator behavior.