fix: add deepseek-v4 models, fix calculate_cost, and improve error ha…

[Comui520]

Summary

• Register deepseek-v4-flash and deepseek-v4-pro in DEEPSEEK_MODELS_DATA
• Fix calculate_cost() across all 8 providers to never return None
• Fix ContextGenerator and Synthesizer to surface errors instead of silently returning empty results

Details

Problem

When pricing data is unknown (model not in registry), calculate_cost() returned None
across ALL providers (DeepSeek, Anthropic, OpenAI, Azure, Gemini, Kimi, Grok, Bedrock),
causing TypeError in ContextGenerator.evaluate_chunk() (total_cost += None).
This error was caught by a broad except Exception and logged but never surfaced,
resulting in empty goldens with no error to the user.

Changes

All 8 provider models — calculate_cost() now returns EvaluationCost(0.0, ...)
when prices are unknown, instead of None. Fixes the Optional[float] return type
to be consistently float.

constants.py — Added model definitions for deepseek-v4-flash and deepseek-v4-pro.

context_generator.py — generate_contexts() and a_generate_contexts() now raise
DeepEvalError when ALL documents fail, instead of silently returning empty contexts.

synthesizer.py — generate_goldens_from_docs() and a_generate_goldens_from_docs()
now raise DeepEvalError when contexts are empty.

Why this matters now

Per DeepSeek API docs, deepseek-chat and
deepseek-reasoner will be deprecated on 2026-07-24 in favor of deepseek-v4-flash
and deepseek-v4-pro.

[vercel]

@Comui520 is attempting to deploy a commit to the Confident AI Team on Vercel.

A member of the Team first needs to authorize it.