Add high-level docs to vortex-btrblocks (#6984)

## Summary

Decided to add some docs to `vortex-btrblocks` as I went through it to
try and understand it better.

## Testing

N/A

Signed-off-by: Connor Tsui <connor.tsui20@gmail.com>

Author: connortsui20

vortex-btrblocks/src/compressor/mod.rs

@@ -1,7 +1,19 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright the Vortex contributors
 
-//! Compressor traits for type-specific compression.
+//! Type-specific compressor traits that drive scheme selection and compression.
+//!
+//! [`Compressor`] defines the interface: generate statistics for an array via
+//! [`Compressor::gen_stats`], and provide available [`Scheme`]s via [`Compressor::schemes`].
+//!
+//! [`CompressorExt`] is blanket-implemented for all `Compressor`s and adds the core logic:
+//!
+//! - [`CompressorExt::choose_scheme`] iterates all schemes, skips excluded ones, and calls
+//!   [`Scheme::expected_compression_ratio`] on each. It returns the scheme with the highest ratio
+//!   above 1.0, or falls back to the default. See the [`scheme`](crate::scheme) module for how
+//!   ratio estimation works.
+//! - [`CompressorExt::compress`] generates stats, calls `choose_scheme()`, and applies the
+//!   result. If compression did not shrink the array, the original is returned.
 
 use vortex_array::ArrayRef;
 use vortex_array::IntoArray;

vortex-btrblocks/src/lib.rs

@@ -18,6 +18,22 @@
 //! - **Statistical Analysis**: Uses data sampling and statistics to predict compression ratios
 //! - **Recursive Structure Handling**: Compresses nested structures like structs and lists
 //!
+//! # How It Works
+//!
+//! [`BtrBlocksCompressor::compress()`] takes an `&ArrayRef` and returns an `ArrayRef` that may
+//! use a different encoding. It first canonicalizes the input, then dispatches by type.
+//! Primitives go to a type-specific `Compressor` (integer, float, or string). Compound types
+//! like structs and lists recurse into their fields and elements.
+//!
+//! Each type-specific compressor holds a static list of `Scheme` implementations (e.g.
+//! BitPacking, ALP, Dict). There is no dynamic registry. The compressor evaluates each scheme by
+//! compressing a ~1% sample and measuring the ratio, then picks the best. See `SchemeExt` for
+//! details on how sampling works.
+//!
+//! Schemes can produce arrays that are themselves further compressed (e.g. FoR then BitPacking),
+//! up to `MAX_CASCADE` (3) layers deep. An `Excludes` set prevents the same scheme from being
+//! applied twice in a chain.
+//!
 //! # Example
 //!
 //! ```rust

vortex-btrblocks/src/scheme.rs

@@ -1,7 +1,24 @@
 // SPDX-License-Identifier: Apache-2.0
 // SPDX-FileCopyrightText: Copyright the Vortex contributors
 
-//! Compression scheme traits.
+//! Compression scheme traits. This is the interface each encoding implements to participate in
+//! compression.
+//!
+//! [`Scheme`] is the core trait. Each encoding (e.g. BitPacking, ALP, Dict) implements it with
+//! two key methods: [`Scheme::expected_compression_ratio`] to estimate how well it compresses
+//! the data, and [`Scheme::compress`] to apply the encoding. Type-specific sub-traits
+//! ([`IntegerScheme`], [`FloatScheme`], [`StringScheme`]) bind schemes to the appropriate stats
+//! and code types.
+//!
+//! [`SchemeExt`] provides the default ratio estimation strategy. It samples ~1% of the array
+//! (minimum [`SAMPLE_SIZE`] values), compresses the sample, and returns the before/after byte
+//! ratio. Schemes can override [`Scheme::expected_compression_ratio`] if they have a cheaper
+//! heuristic.
+//!
+//! [`IntegerScheme`]: crate::compressor::integer::IntegerScheme
+//! [`FloatScheme`]: crate::compressor::float::FloatScheme
+//! [`StringScheme`]: crate::compressor::string::StringScheme
+//! [`SAMPLE_SIZE`]: crate::stats::SAMPLE_SIZE
 
 use std::fmt::Debug;
 use std::hash::Hash;