Add config tests for DeepSeek V3 architecture adapter

[SanjidMzi]

Part of #1302.

This pr adds unit tests for DeepSeekV3ArchitectureAdapter covering the config flags it
sets, the component-mapping structure and HF module paths, the MLA attention
submodule mapping, and the (empty) weight-conversion set.

V3 differs from V2 at the unit level: V3 always compresses Q via
the two-stage LoRA path (that's why there is no direct q_proj fallback and every
attention projection is required), q_a_layernorm is an RMSNorm rather than a
plain component, and the router gate is bridged as an optional component.

Behavioural coverage (forward pass, hook firing) already lives in
tests/integration/model_bridge/test_deepseek_adapter.py; these tests
complement it at the unit tier, following the same pattern as the existing
test_deepseek_v2_adapter.py.

make format, uv run mypy ., and make unit-test all pass locally.

[jlarson4]

Looks great! Thank you for putting this together @SanjidMzi!