feat(nn): add WithBias option for Linear layer

[kolkov]

Summary

Add functional options pattern to NewLinear for optional bias configuration.

• Add LinearOption type and WithBias(bool) function
• Add HasBias() method for introspection
• Update SwiGLUFFN to use public API (remove internal newLinearNoBias)
• Export WithBias in public nn package

Usage

// With bias (default, backwards compatible)
layer := nn.NewLinear(784, 128, backend)

// Without bias (for LLaMA-style models, LM head, etc.)
lmHead := nn.NewLinear(hiddenSize, vocabSize, backend, nn.WithBias(false))

Motivation

Many neural network architectures require Linear layers without bias:

• LM Head in language models (GPT, LLaMA, HRM)
• Attention projections (some architectures)
• SwiGLU FFN (already used internally)

Test plan

• TestLinear_WithBias - tests default, explicit true, and false cases
• TestLinear_NoBias_Forward - tests forward pass without bias
• TestLinear_NoBias_StateDict - tests state dict without bias
• All existing tests pass
• golangci-lint: 0 issues
• Pre-release check passed

[codecov]

Codecov Report

❌ Patch coverage is 84.00000% with 4 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
nn/nn.go	0.00%	4 Missing ⚠️

📢 Thoughts on this report? Let us know!