feat: add experimental onpair string encoding

Cargo.toml

@@ -54,6 +54,8 @@ members = [
     "encodings/zstd",
     "encodings/bytebool",
     "encodings/parquet-variant",
+    # Experimental encodings
+    "encodings/experimental/onpair",
     # Benchmarks
     "benchmarks/lance-bench",
     "benchmarks/compress-bench",
@@ -188,6 +190,7 @@ num_enum = { version = "0.7.3", default-features = false }
 object_store = { version = "0.13.1", default-features = false }
 once_cell = "1.21"
 oneshot = { version = "0.2.0", features = ["async"] }
+onpair = { version = "0.0.3", git = "https://github.com/spiraldb/onpair.git", rev = "53e8ca6081d377e9933d999cef286e26bf52e2c7" }
 opentelemetry = "0.32.0"
 opentelemetry-otlp = "0.32.0"
 opentelemetry_sdk = "0.32.0"
@@ -288,6 +291,7 @@ vortex-ipc = { version = "0.1.0", path = "./vortex-ipc", default-features = fals
 vortex-layout = { version = "0.1.0", path = "./vortex-layout", default-features = false }
 vortex-mask = { version = "0.1.0", path = "./vortex-mask", default-features = false }
 vortex-metrics = { version = "0.1.0", path = "./vortex-metrics", default-features = false }
+vortex-onpair = { version = "0.1.0", path = "./encodings/experimental/onpair", default-features = false }
 vortex-parquet-variant = { version = "0.1.0", path = "./encodings/parquet-variant" }
 vortex-pco = { version = "0.1.0", path = "./encodings/pco", default-features = false }
 vortex-proto = { version = "0.1.0", path = "./vortex-proto", default-features = false }

encodings/experimental/onpair/Cargo.toml

@@ -0,0 +1,40 @@
+[package]
+name = "vortex-onpair"
+authors = { workspace = true }
+categories = { workspace = true }
+description = "Vortex OnPair string array encoding (dict-12, pushdown predicates)"
+edition = { workspace = true }
+homepage = { workspace = true }
+include = { workspace = true }
+keywords = { workspace = true }
+license = { workspace = true }
+readme = "README.md"
+repository = { workspace = true }
+rust-version = { workspace = true }
+version = { workspace = true }
+
+[lints]
+workspace = true
+
+[dependencies]
+memchr = { workspace = true }
+num-traits = { workspace = true }
+onpair = { workspace = true }
+prost = { workspace = true }
+vortex-array = { workspace = true }
+vortex-buffer = { workspace = true }
+vortex-error = { workspace = true }
+vortex-mask = { workspace = true }
+vortex-session = { workspace = true }
+
+[features]
+_test-harness = ["vortex-array/_test-harness"]
+
+[dev-dependencies]
+divan = { workspace = true }
+rstest = { workspace = true }
+vortex-array = { workspace = true, features = ["_test-harness"] }
+
+[[bench]]
+name = "decode"
+harness = false

encodings/experimental/onpair/README.md

@@ -0,0 +1,39 @@
+# Vortex OnPair
+
+A Vortex Encoding for Binary and Utf8 data that uses the
+[OnPair][onpair] short-string compression algorithm. OnPair is a
+dictionary-based encoder with fast per-row random access.
+
+The trainer / encoder lives in the standalone [`onpair`][onpair-crate]
+crate; this crate wraps the resulting column as a Vortex array with
+cascading-compressor support on every integer child.
+
+## Compute
+
+Like the FSST encoding, this crate provides `cast` and `filter`
+pushdown. Other operators fall back to ordinary decompression.
+
+## Default Configuration
+
+The default training preset is **dict-12**: 12 bits per token,
+dictionary capped at 4 096 entries. Token codes are stored as a
+`PrimitiveArray<u16>`; downstream `FastLanes::BitPacking` losslessly
+narrows the child to exactly `bits`-bit codes on disk.
+
+## Layout
+
+- Buffer 0 — `dict_bytes`: dictionary blob built by the OnPair trainer,
+  padded with `MAX_TOKEN_SIZE` trailing zero bytes so the over-copy
+  decoder can read 16 bytes past the last token.
+- Slot 0 — `dict_offsets`: `PrimitiveArray<u32>`, len `dict_size + 1`.
+- Slot 1 — `codes`: `PrimitiveArray<u16>`, length `total_tokens`.
+- Slot 2 — `codes_offsets`: `PrimitiveArray<u32>`, length `num_rows + 1`.
+- Slot 3 — `uncompressed_lengths`: integer `PrimitiveArray`, length
+  `num_rows`.
+- Slot 4 — optional validity child.
+
+All four integer slot children flow through the standard cascading
+compressor pipeline (FoR / BitPacking / RunEnd / etc.).
+
+[onpair]: https://arxiv.org/abs/2508.02280
+[onpair-crate]: https://github.com/spiraldb/onpair