docs: reciprocal link to the type-family map + vocabulary alignment (rate limits vs resource/occupancy grades)

[arena-ai-coding-agent]

The estate's type-research families share a vocabulary + boundary map in nextgen-typing/docs/TYPE-CONNECTIONS.adoc (tracked by nextgen-typing#118). Typell sits in the map's "Main Chain" (katagoria → typell → typed-wasm → PanLL) with the boundary "a conceptual connection in this guide does not establish that any family is integrated into these projects". Two small coherence asks:

1. Reciprocal link + IS-NOT. The README's "What Typell is NOT" section is good boundary practice; consider adding one line: typell is the verification kernel — it is not one of the research families (Echo/Epistemic/Tropical/Residual/Choreographic/Occupancy), and consumption of any family's concepts is a decision documented here, not implied by the map. Link TYPE-CONNECTIONS.adoc.
2. Vocabulary alignment. The README claims the kernel ensures "resources are used exactly once, transactions are atomic, and rate limits are respected". In the shared glossary, budget/credit-like quantities are resource grades (and, for live-state kinds, occupancy grades — see the cost/state split being added to the map under #118). Worth naming which one "rate limits" means (declared constant bound? state credits? adaptive?), since the family map's whole point is that cost bounds and state bounds compose differently. Same for "connections close" — the session discipline vocabulary is shared with occupancy-types' session IR (close = explicit, consuming).
3. Coverage table ≠ family map. The "Target type-safety coverage" table (dependent/linear/session/quantitative/effect/modal) is a kernel feature matrix; a footnote that it is not the 10-levels ladder (kategoria) nor the family map would close the loop on all three taxonomies.

Tracked by: hyperpolymath/nextgen-typing#118