docs, add AI skill knowledge base for h5cpp usage patterns

[steven-varga]

Problem

AI coding assistants hallucinate h5cpp usage because the library is niche,
template-heavy, and has non-obvious invariants: include order matters, property
lists have type constraints, and type dispatch goes through traits that are
invisible at the call site. There is no curated, machine-readable corpus of
verified h5cpp patterns for models to draw on.

Goal

A self-maintaining skill knowledge base anchored to the test suite, structured
for injection into all four agents (Claude, Kimi, Codex, Copilot).

Architecture

docs/skills/ — verified example corpus

Structured Markdown files, one per topic, with YAML front-matter:

---
skill: io-write-vector
tags: [io, write, stl, vector]
hdf5_min: 1.10
tested: true   # snippet appears verbatim in a ctest
---

Constraint: every snippet tagged tested: true must appear verbatim in a
ctest. A CI script scripts/verify-skills diffs skill snippets against their
test source and fails on divergence.

docs/skills/index.md — tag index

Maps tags → skill files for agent retrieval.

scripts/mem integration

Skill files imported into knowledge.db as observations on a h5cpp-skills
entity. mem-consolidate re-imports on each daily run.

Scope (this issue — core I/O patterns)

• docs/skills/index.md
• docs/skills/includes.md — entry points and include order
• docs/skills/io-create.md — h5::create dataset and file patterns
• docs/skills/io-write.md — h5::write full and partial
• docs/skills/io-read.md — h5::read full and partial
• docs/skills/io-append.md — h5::append / packet table streaming
• docs/skills/properties.md — property list composition with operator|
• scripts/verify-skills — CI verification stub

Out of scope (follow-on issues)

• Compound type / struct patterns
• Attribute I/O patterns
• Linalg adapter patterns (Armadillo, Eigen, xtensor, …)
• MPI collective / independent patterns
• S3 / ROS3 patterns (covered in feature, add HDF5 S3 / ROS3 VFD support #188)
• Fine-tuning dataset generation