docs: polished README + CMake build + MIT license

[ldocs17]

What

Documentation overhaul for the CSV→Parquet engine.

• README rewritten: one-line pitch, Mermaid architecture diagram, accurate "how it works" (zero-copy mmap, lock-free chunk parallelism, predictive Arrow allocation), platform build instructions, benchmark methodology, project structure, and a roadmap.
• CMakeLists.txt added so the README's cmake build steps actually work (finds Arrow + Parquet, links pthreads, handles new/old Arrow target names).
• LICENSE (MIT) added — swap if you prefer a different license.

Notes / honest caveats

• The README documents the current behavior truthfully: input path, output path, and thread count are constants at the top of CSVtoParquet.cpp. CLI args are listed as the first roadmap item, not claimed as existing.
• No CI workflow included (kept this docs-only), so the build-status badge was intentionally omitted to avoid a broken badge. CI is on the roadmap.
• Benchmark numbers are carried over from the existing README/commit history — re-verify before quoting them elsewhere.

🤖 Generated with Claude Code

[ldocs17]

Added a GitHub Actions build workflow (.github/workflows/ci.yml) to this PR — it installs Arrow + Parquet via conda-forge libarrow-all and builds with the new CMakeLists on every push/PR. First run passed in 43s ✅, so the build badge (now restored in the README) is green. Only non-fatal deprecation warnings remain (Node 20 / auto-activate-base).