fix(runtime): typed-fault train follow-ups

[mfw78]

Summary

Closes #206; part of the typed-fault train stacked on the epic-80 stack, merges in stack order.

This follows up on the 15-item list drawn up when the host-error teardown was still in flight.
Most of it turned out to already be done or superseded by the base branch, feat/fault-teardown, which by the time this branch was cut had deleted HostError/host-error/host-error-kind repo-wide, moved chain, cow-api, generic-interfaces and module-exports onto typed errors over the shared Fault vocabulary, and swept the whole prose and diagram tree to match. Two items were genuinely live work; this PR implements them.

Changes

• docs(host-chain): rustdoc on request_batch (crates/nexum-runtime/src/host/impls/chain.rs) stating that the outer ChainError is reserved for a whole-batch failure, which this host never produces, versus a per-entry RpcResult::Err. Comment-only, no behaviour change.
• fix(cow-sdk): crates/shepherd-sdk/src/cow/app_data.rs folds a length-capped, char-boundary-safe, ellipsised prefix of the raw orderbook response body into the Fault::Internal shape-error message via a new truncate_for_log helper, restoring the diagnostic the old data: Some(response) field used to carry. Updated the renamed test and added a new oversized-body truncation test.

Dispositions of the remaining follow-ups

• Already done in feat/fault-teardown: the From/TryFrom wit-bindgen adapter, consumer migration off HostError, the 0.2 error-model docs reconciliation, migrating chain/cow-api/exports onto typed errors, retiring the HostError bridge, the prose rewrite across docs/00-08 and the migration guide, and the ADR-0009 superseded note pointing at ADR-0011.
• Confirmed non-issue: the duplicated_attributes clippy lint flagged against nexum-sdk does not surface under the pinned 1.94 toolchain; the crate is clean under -D warnings.
• Moot: cow-api docs, ADRs and .mmd diagram sources carry no stale host-error references and no rendered .png files exist to regenerate.
• Deferred, out of scope for this PR: richer fault cases for the remote-store, identity and messaging backends, which have not landed yet and still report Unsupported placeholders; and wiring retry_after_ms from an alloy backoff hint, which is blocked because RpcErrorExt::backoff_hint is pub(crate) in alloy-transport and TransportErrorKind::HttpError does not carry response headers, so there is no Retry-After value to read without reimplementing alloy's private parsing. That gap also blocks a RetryAction::Backoff producer, since there is no populated retry_after_ms to derive one from. Optional backoff/URL fields in payload-free logs are accepted as-is.

Test plan

• cargo fmt --all --check
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo test -p shepherd-sdk --all-features (28 unit tests, 3 doctests)
• No WIT, boot or dispatch paths touched, so no e2e wasm build required; the nexum-runtime change is a fenceless rustdoc comment validated by the clippy compile.

Notes for reviewer

This is a small, mostly-dispositional PR: two real fixes plus a written-down accounting of why the other thirteen follow-ups are either already covered by feat/fault-teardown or still genuinely deferred (and why). The alloy backoff_hint limitation is worth a second look if anyone knows of a public alternative; absent that, treat the RetryAction::Backoff producer as blocked rather than skipped.

AI Assistance: Claude Code (Opus 4.8 implementation and adversarial review, Sonnet 5 PR authoring) used for the full change.