[Wave 4] Structured GP and Uncertainty-Aware NN Methods

## Goal
Add the structured and scalable methods that make `pyrox` distinctive: temporal/state-space GPs, pathwise posterior sampling, multi-output GP structure, and uncertainty-aware NN composition patterns.

## Why This Wave Exists
By Wave 4, the package should move beyond dense/sparse basics into the structured algorithms that motivate the combined `gp` and `nn` design.

## Wave / Milestone
- Wave: `wave:4`
- Milestone: `v0.4-structured-gp-and-uncertainty-aware-nn`

## Canonical Epics
- [ ] #34 [Epic] 4.A GP Temporal and State-Space Methods
- [ ] #35 [Epic] 4.B GP Pathwise Sampling and Multi-Output Models
- [ ] #36 [Epic] 4.C NN Uncertainty Patterns

## Sequential Dependencies
- `#34` and `#35` can run in parallel once Wave 3 GP work is stable.
- `#36` can run in parallel once Wave 3 NN work is stable.
- No single theme epic should absorb the others’ scope; keep temporal, pathwise/multi-output, and NN uncertainty separate.

## Definition of Done
- Temporal/state-space GPs are represented via SDE kernels and Markov workflows.
- Pathwise posterior sampling and multi-output GP structure are present.
- NN uncertainty patterns cover richer spectral features plus spectral-normalized distance-aware models.

## Relationships
- Blocked by #24.
- Blocks #43.