docs: extend CLAUDE.md with migrated agent notes

[rubenhensen]

Extends the existing CLAUDE.md with agent notes migrated from the private dobby memory repo (encryption4all/dobby, formerly repos/<this-repo>/), verified against the current state of the code before migration.

Going forward this file is the single home for durable repo knowledge: dobby updates it in the same PR whenever it learns something worth keeping.

[dobby-coder]

Thanks for consolidating the repo knowledge into CLAUDE.md — the CI/release/overview sections are accurate and the format is exactly right for durable notes.

One blocking issue: the entire dependency section (lines 30–58) documents the pre-migration world as if it were current, even though the PR body says it was "verified against the current state of the code." The curve25519-dalek 5.0 / rand 0.10 / sha3 0.12 migration already landed in main via #24 (commit 0ebff0e, now HEAD). I verified against this branch: Cargo.toml pins curve25519-dalek =5.0.0-rc.1, rand_core 0.10, sha3 0.12, shake 0.1, rand 0.10; OsRng appears nowhere in src/; and the API is pub fn setup<R: CryptoRng> + Scalar::random(r) with rand::rng().

So a future agent reading this file would re-investigate a migration that's done and trust dep versions that are two states stale. The individual migration facts (shake crate split, rand::rng(), CryptoRng: Rng) are correct and match the code — they just need to be reframed from "when this moves" to "this moved (#24)". Please rewrite the dependency section to describe the current pinned state as landed, then this is good to merge. Inline notes below.

[rubenhensen]

/dobby fix all comments you made

[dobby-coder]

The prior CHANGES_REQUESTED flagged the whole dependency section as documenting the pre-migration world. The reframe has landed for the dalek/rand/sha3/shake pins (now correctly described as landed via #24, with the rc-1 caveat) — I verified those against Cargo.toml and src/gg.rs on HEAD and they are accurate.

One issue from that same defect is still present: the bincode tombstone (line 39) documents pre-migration state. Cargo.toml pins bincode-next = "=3.0.0-rc.13" and src/gg.rs uses bincode_next:: — commit fdc9bac already migrated off the original unmaintained bincode crate. So "stay on bincode 2.x … do not bump to bincode 3.0.0" is false and would mislead a future agent. Fix inline and this is good to merge.

[dobby-coder]

Clean sign-off. Both prior blocking findings are now resolved on HEAD:

• Dependency section (cycle 1): reframed from "pre-migration / when this moves" to the current landed state — the curve25519-dalek 5.0 / rand 0.10 / sha3 0.12 / shake 0.1 pins are now described as landed via #24, with the =5.0.0-rc.1 release-candidate caveat retained.
• bincode tombstone (cycle 2): now correctly states tests use the bincode-next fork (=3.0.0-rc.13, gg.rs calls bincode_next::) migrated off the original unmaintained bincode in fdc9bac, with the tombstone pointing the right way ("do not switch back").

Per-rule sweep on the applicable rules (conventional-commit PR title, docs style / no em dashes / banned words, no-justification-paragraphs) is clean. Title docs: extend CLAUDE.md with migrated agent notes is a valid conventional commit. No security advisory referenced. Good to merge.