docs: shared in-memory ABI (plain interchange form)

[gargiulofrancesco]

Summary

Defines the shared in-memory format for an OnPair-compressed column so that independent implementations are interoperable: a column produced by one is intelligible to another.

It specifies the buffer layout (dictionary bytes + offsets, code stream, optional row offsets), the invariants a decoder relies on (256-token completeness, offset monotonicity, dictionary read-padding, every code < N), and the language-neutral structures for addressing a column in memory.

Scope: this is the plain interchange form (codes are plain u16). Compressed code encodings (bit-packing, etc.) and on-disk serialization are out of scope and remain each implementation's own concern — this changes only the shared exchange contract, not how anyone serializes or stores data internally.

Replaces the previous docs/binary-format.md.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[codspeed-hq]

Merging this PR will not alter performance

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 2 improved benchmarks
❌ 3 regressed benchmarks
✅ 27 untouched benchmarks
⏩ 2 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	WallTime	decompress_all[12]	753.6 µs	891.2 µs	-15.44%
❌	WallTime	decompress_all[16]	775.1 µs	888.4 µs	-12.75%
❌	WallTime	decompress_all[("o_comment", 12)]	15.2 ms	17.1 ms	-10.66%
⚡	WallTime	train_and_compress[("l_comment", 12)]	469.9 ms	422.6 ms	+11.2%
⚡	WallTime	train_and_compress[("o_comment", 12)]	392.1 ms	355.6 ms	+10.25%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing docs/in-memory-abi (1766c09) with develop (cb4ea96)}

Footnotes

1. 2 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[joseph-isaacs]

You can do this either before or in decompression

[gargiulofrancesco]

I reworked §6 to list what must hold for a conformant column without prescribing when checks must occur. I also made the list comprehensive (e.g. dictionary token uniqueness, which was missing)