[Epic] 5.C NN Advanced Layer Families

## Theme
Deliver the advanced NN layers and tutorials that round out the `pyrox.nn` story after the foundational and Wave 4 uncertainty patterns are in place.

## Parent Wave
- Wave epic: `#43`
- Wave label: `wave:5-advanced`
- Milestone: `v0.5-advanced-algorithms-and-integrations`

## Motivation
These families close the gap with Edward2 / TFP-layers for users migrating from TensorFlow: multi-modal regression heads, structured sequence output, and the convolutional / recurrent / attention families that make `pyrox.nn` usable beyond MLP backbones. After the pyrox/geonnax split, the deterministic forward kernels live upstream in geonnax; this epic tracks only the Bayesian wrappers and the family build-out.

## Issues
- [ ] #51 — Edward2-style advanced output heads (`MixtureOfGaussiansDenseFA`, `LinearChainCRF`)
- [ ] #52 — Bayesian convolutional / recurrent / attention layer families

## Execution Notes
#51 is blocked by jejjohnson/geonnax#12 (deterministic MoG-head + CRF cores). #52 is independent of #51 and can start anytime; its masterclass docs/examples should land in the same PRs as the layers.

## Parallelism
- Can run in parallel with: #44 and #45
- Blocked by (inside this wave): none
- Must complete before: final advanced-wave polish and cross-cutting docs

## Definition of Done
- Advanced uncertainty layers and sequence/vision families are represented with at least one credible example path.

## Relationships
- Parent wave: #43.