Record the namespace convention's real state, and the family's open divergences

[tap]

What this changes

Updates the namespace-convention section of README.md to say what is actually true across the family, and adds a Known divergences section recording the remaining unevenness with its known fix and the reason it has not been taken.

Why

The convention table claimed a tap::<library> CMake alias for every library, but only DspTap and RatioTap had ever created one. A table that documents a target as if it were current state is worse than no table: five repos looked compliant and were not. The companion PRs in AmbiTap, MuTap, SampleRateTap, TapTools and OscTap add the missing aliases, so the table can now be a description rather than an aspiration.

The rest of the unevenness found during the audit had no home. Written into a section here, it is decided; left in a session transcript, it gets rediscovered.

Verification

Documentation only — no build, and nothing here is consumed at build time or drift-checked (README.md is not in scripts/sync.sh's copy set, so no consumer re-sync is implied).

Every claim in the new text was checked against the repos rather than recalled:

• tap::<library> alias now present in all seven libraries — grepped add_library(tap::… ALIAS …) across each CMakeLists.txt.
• Header paths: tap/dsp/ and tap/ratio/ comply; ambitap/, srt/, mutap/, taptools/, osctap/ do not.
• Namespaces already comply everywhere (tap::ambi, tap::dsp, tap::mu, tap::osc, tap::ratio, tap::samplerate, tap::tools).
• install/export: 8 install() + config package in AmbiTap, 6 in TapTools, zero in DspTap/MuTap/RatioTap/SampleRateTap.
• Option prefixes: six distinct forms across six libraries.
• PythonTap is a host package, not a library — no include/, one Max external (MIN_EXTERNAL(python)), min-api submodule.

Notes for the reviewer

• Canonical TapHouse file changed — but only README.md, which is deliberately outside the guarded set, so this needs no tag and no consumer re-sync. Merge order still matters for the reading of it: the ten companion PRs are what make the alias column true.
• The table gained a Header path column precisely because that is the column still lagging, and for a real reason: renaming include/mutap/x.h → include/tap/mu/x.h breaks every #include in every consumer. It stays a migrate-when-touched item, with a forwarding header during transition.
• Two items in the divergences list are recorded as deliberate non-actions, not TODOs: install/export rules (all four repos lacking them are consumed via add_subdirectory, so the rules would be dead CMake) and the Catch2-vs-GoogleTest split (a real difference in kind — converting either way would discard working tests).
• Two are recorded as blocked on a machine I do not have: PythonTap's C++17 targets (its build needs an embedded Python runtime whose installer is macOS/Windows only) and the macOS platform-string spelling in the Max packages (arm64 vs aarch64 — unverifiable without Max).

---

Generated by Claude Code