refactor(core): thin the CLI — move parsing/resolution into ant-core, route all daemon calls through the client

[Nic-dorman]

Summary

Five mechanical fixes from the CLI-architecture audit (the "CLI should be a thin adapter" contract in CLAUDE.md), one commit per finding:

• PortRange parsing (audit docs: comprehensive API documentation for data, wallet, devnet, and library usage #7): impl FromStr for PortRange in ant-core next to the type; CLI uses .parse() instead of its own parse_port_range.
• Env vars (audit fix: resolve HRTB lifetime issue in stream closures #8): AddNodeOpts::parse_env_vars owns KEY=VALUE parsing; CLI helper removed.
• Bootstrap resolution (audit fix: migrate NAT-friendly quote selection and process exit fix #4): explicit > devnet manifest > bootstrap_peers.toml priority (with manifest-peer filtering) moved to ant_core::config::resolve_bootstrap_peers; no-source case is now typed Error::NoBootstrapPeers, and a selected-but-empty/fully-filtered manifest errors rather than silently falling back.
• Daemon calls (audit Migrate from saorsa-node to ant-node 0.6 #3): node add/node reset no longer hand-roll reqwest against the daemon — new daemon::client::add_node / client::reset matching every other endpoint.
• Registry race (audit Unify all clients into ant-client #2): node start/stop --service-name resolved the node ID by reading node_registry.json directly while the daemon (the registry owner) was running. New client::resolve_node_id_by_name goes through GET /nodes/status; the CLI no longer touches the registry file at all.
• Progress reporters (audit chore: bump version to 0.1.1 #9): the two near-identical ProgressReporter impls (CliProgress in add.rs, CliUpdateProgress in update.rs) collapsed into one progress::CliProgress on stderr. Side fix: ant node add --json no longer interleaves download progress with the JSON result (NoopProgress in JSON mode, matching ant update).

Linear issue

V2-189

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Compatibility

• Wire: none
• Storage: none
• API: additive — ant-core gains config::resolve_bootstrap_peers, daemon::client::{add_node, reset, resolve_node_id_by_name}, impl FromStr for PortRange, AddNodeOpts::parse_env_vars, and Error::NoBootstrapPeers; nothing is removed. CLI: node add --json download progress moves stdout→stderr, and daemon add/reset error surfacing changes to the shared HTTP request error: {body} form.

Semver impact

• breaking
• feature
• fix

Test evidence

• cargo test --workspace: 446 passed — new unit tests for PortRange::from_str, parse_env_vars, and resolve_bootstrap_peers priority, including the empty-manifest and fully-filtered-manifest NoBootstrapPeers cases.
• clippy -D warnings clean, fmt clean.
• Live smoke against a running daemon (empty registry): name-miss and name-hit resolution for start/stop, reset --force --json no-op, add --json full round-trip with a stub binary, and client-side parse failures for an inverted port range and a malformed --env — all outputs byte-compatible with the old CLI.
• Rebased onto main 2026-08-13; re-verified fmt / cargo check --workspace / ant-core lib tests locally on the rebased head.

New dependency

none — this PR removes ant-cli's direct reqwest dependency (the moved daemon calls now go through ant_core::node::daemon::client).

ADR

n/a (Tier 1).

Mitigation / rollback

Revert the PR. The change is client-only with no persisted state, wire-protocol, or stored-format impact, so backing it out restores the previous CLI behavior wholesale.

Behavior notes

• Parse/lookup error texts are preserved (No node found with service name '…', Expected KEY=VALUE, port-range messages near-identical).
• Daemon-error surfacing for add/reset changes from Daemon returned error: {body} to HTTP request error: {body} (the shared daemon-client convention).
• node add download progress moves stdout → stderr, consistent with the rest of the progress UI.

[dirvine]

Reviewed current head c8b032b10b63bd4f7a0c0497e4f4f4a1bdbd2ae8. The CLI/core moves, daemon routes, name resolution and JSON progress handling look sound, and all 12 CI checks are green. I found two items to address before merge:

1. Remove unrelated generated/vendor artifacts. This PR adds vendor/ant-node/Cargo.toml.orig, vendor/saorsa-core/.cache/saorsa/bootstrap_cache.json, and vendor/saorsa-core/.cache/saorsa/bootstrap_cache.lock. The cache includes a machine/runtime instance ID and timestamp; none of these files belong to the CLI-thinning change. Please remove them and consider ignoring nested .cache/ and *.orig artifacts.

2. Make resolve_bootstrap_peers uphold its new typed-error contract. ant-core/src/config.rs:90-93 says NoBootstrapPeers is returned when no source yields peers, but lines 102-108 return Ok(vec![]) for an empty manifest or when every manifest MultiAddr is filtered out. Because selecting a manifest also disables the public peer cache, the CLI can start apparently successfully with zero peers and fail only during the data operation. Please return Error::NoBootstrapPeers for an empty filtered manifest result and add tests for an empty manifest and a fully filtered manifest. Do not fall back to the public config when a manifest was explicitly selected.

The empty-manifest behaviour existed in the old CLI helper, so this is not a newly introduced runtime regression; it is a mismatch in the new public ant-core API and misses the purpose of the typed error.

Verified locally: cargo fmt --all -- --check, focused node::types and config tests, cargo check --all-targets; independent review also ran the ant-core library and daemon integration tests successfully.

[Nic-dorman]

Both review items addressed in ec6d5d5:

1. Vendor artifacts removed — vendor/ant-node/Cargo.toml.orig and the vendor/saorsa-core/.cache/saorsa/bootstrap_cache.{json,lock} files are deleted from the branch, and .gitignore now covers nested .cache/ directories and *.orig files so they can't come back.

2. resolve_bootstrap_peers contract fixed — a selected manifest that is empty, or whose entries are all filtered out (no resolvable socket address), now returns Error::NoBootstrapPeers instead of Ok([]), with no fallback to the public config. The CLI's only call site propagates the error, so startup fails immediately rather than at the first data operation. Added resolve_bootstrap_errors_on_empty_manifest and resolve_bootstrap_errors_when_all_manifest_peers_filtered (the latter uses a Bluetooth MultiAddr whose socket_addr() is None), and tightened the doc comment to say a selected manifest is authoritative.

cargo fmt --check, cargo clippy --all-targets --all-features -D warnings, and the ant-core test suite all pass locally.

[dirvine]

Re-reviewed current head ec6d5d5d7992cd1a1f2296dec26568137b3d1076 against the prior findings. Both are addressed:

• The stray vendor .orig/.cache/lock artifacts are removed from the tracked tree, and the ignore rules cover the generated artifact shapes without ignoring vendor/ wholesale.
• A selected manifest is now authoritative: empty or fully filtered manifest peers return Error::NoBootstrapPeers before node startup, with no public-config or peer-cache fallback. The new empty-manifest and non-socket MultiAddr regressions exercise both paths.

No new code blocker found. Verified locally on this exact head: cargo fmt --all -- --check, four focused bootstrap resolver tests, and cargo check --all-targets all pass. Independent adversarial probes also confirmed explicit peers retain priority and an empty manifest cannot fall back to configured public peers.

Current CI is 9/12 green with no failures; macOS/Ubuntu E2E and Ubuntu Merkle E2E are still in progress. Code review is clear; merge readiness remains conditional on those jobs finishing green.