docs: refine documentation tone, eliminate ai artifacts, and expand technical depth

[MichalRedm]

Summary

• Refactored the entire documentation suite to eliminate AI-generated artifacts, decorative emojis in headings/bullets, promotional boilerplate, and superficial placeholders.
• Expanded user guides across all framework analytical layers (Data Layer, Model Layer, Decision Boundary Analysis, Feature Importance, Clustering Dynamics, Recurring Concepts, and Dashboard) into comprehensive, publication-grade academic prose with formal mathematical formulations.
• Upgraded the documentation landing page (docs/index.md) to reflect the published ECML PKDD 2026 paper, Springer chapter citation, NCN grant funding, and three-layer synchronous processing architecture.
• Enriched all 10 API reference pages under docs/api/ with module architectural overviews, key class/function matrices, and standard import workflows preceding the autogenerated mkdocstrings directives.

Motivation

Following pull request #22 which established the MkDocs infrastructure and autogenerated API reference, the documentation retained visible signs of AI-generated content (excessive emojis in headings, shallow 30-line skeleton pages, formulaic bullet lists, and conversational filler). Drawing inspiration from the evolution of README.md—which transitioned from an emoji-heavy draft to a rigorous, academic presentation for ECML PKDD 2026—this PR elevates the documentation to match the project's peer-reviewed research caliber.

Key Changes

1. Landing Page & Quickstart

• Replaced emoji headings (🌟 Overview, 🏛️ Architecture, 🚀 Quick Navigation) and centered HTML banner in docs/index.md with an academic header, publication badges, formal problem decomposition, three-layer architectural narrative, and clean navigation tables.
• Overhauled docs/getting-started/quickstart.md to present a cohesive, realistic end-to-end stream forensics workflow (Hyperplane drift generation $\to$ statistical two-sample tests $\to$ sequential error tracking with CUSUM lookback $\to$ temporal discriminator feature attribution $\to$ dashboard launch).
• Expanded docs/getting-started/installation.md with explicit virtual environment setup and a detailed capability extras table.

2. Analytical User Guides

• Data Layer (docs/user-guide/data-layer.md): Added formal mathematical formulations for Kolmogorov-Smirnov ($D_{KS}$), Anderson-Darling ($A^2$), and 1D Wasserstein distance ($\mathcal{W}_1$), along with practical guidance on window size selection and multiple hypothesis testing.
• Model Layer (docs/user-guide/model-layer.md): Detailed River's DDM Bernoulli process formulation ($p_t \pm 2s/3s$ thresholds), CUSUM/gradient lookback onset localization mechanics, and the DriftDescription dataclass schema.
• Decision Boundaries (docs/user-guide/boundary-analysis.md): Explained why standard dimensionality reduction lacks continuous inverse mapping ($\mathbb{R}^2 \to \mathbb{R}^D$), detailed SSNP's composite reconstruction and neighborhood preservation loss, and documented disagreement decision tree rule extraction.
• Feature Importance (docs/user-guide/feature-importance.md): Formalized the temporal discriminator paradigm, explained the distinction between data drift ($Z = X$) and real concept drift ($Z = [X, Y]$), and documented conformal selective prediction with abstention.
• Clustering Dynamics (docs/user-guide/clustering-dynamics.md): Documented dynamic $K$ selection via X-means BIC score, formulated centroid matching as optimal bipartite assignment via the Hungarian algorithm ($O(K^3)$), and detailed cluster velocity, birth, and death metrics.
• Recurring Concepts (docs/user-guide/recurring-concepts.md): Formalized prototype tree representations (ProTree), pairwise distance matrices ("minimal_distance", "mean_distance"), and HDBSCAN regime clustering with outlier correction.
• Interactive Dashboard (docs/user-guide/dashboard.md): Detailed the decoupled architecture, session state isolation, computation gating, full tab-by-tab workflow, and Shaker dataset stitcher dialog.

3. Theory & Developer Standards

• Drift Taxonomy (docs/theory/drift-taxonomy.md): Formalized Bayesian decomposition $P(X, Y) = P(X)P(Y \mid X) = P(Y)P(X \mid Y)$, temporal velocity profiles (abrupt, gradual sigmoidal mixture, incremental, recurring), and mapped them to STRIDE's three layers with formal citations.
• Mathematical Formulations (docs/theory/mathematical-formulations.md): Comprehensive mathematical reference including Kantorovich-Rubinstein duality, asymptotic ECDF distributions, SSNP loss formulation, and conformal prediction guarantees.
• Contributing & Agent Standards (docs/developer/): Outlined architectural invariants (pure core, headless plotting), Git conventional commit protocols, and codified Google-style docstring invariants with type deduplication rules.

4. API Reference Manuals

• Enriched all 10 API reference documents (api/datasets.md, api/drift.md, api/exceptions.md, api/models.md, api/plotting.md, api/xai-boundary.md, api/xai-clustering.md, api/xai-importance.md, api/xai-recurrence.md, api/xai-stats.md) with module architectural contexts and summary tables of exported classes, exceptions, and methods.

Verification

• ruff check . passed with 0 errors (All checks passed!)
• ruff format --check . passed cleanly with 0 formatting warnings (113 files already formatted)
• python -m unittest discover tests passed with 42/42 tests passing (OK)
• mkdocs build --strict built documentation with 0 warnings or errors in strict mode
• Verified zero remaining emojis across all markdown files in docs/