feat: throttle outgoing bandwidth and redirect saturated serves to peers

[MichaelTaylor3d]

Summary

• Adds a configurable outgoing-bandwidth cap (DIG_NODE_MAX_OUTGOING_BYTES_PER_SEC, unlimited by default) on the node's serve path (new bandwidth module, OutgoingThrottle: a fixed-1s-window byte budget).
• When serving would push the node's outgoing traffic over the cap, it resolves alternate holders via the DHT (download::NodeContent::find_providers) and answers with the same CONTENT_REDIRECT (-32008) shape + hop-cap discipline the existing #165 redirect-on-miss mechanism already uses — extending redirect from "not held" to "held but saturated," per dig_ecosystem issue #30.
• Graceful fallback: when no alternate holder is known (or no P2P engine is attached — the in-process FFI/DIG-Browser path), the node serves the request anyway. The throttle changes where a request is served from when it can, never whether — nothing is dropped or fails closed.
• Wired into the three surfaces that return locally-held resource bytes: dig.getContent's local-first serve (cold cache hit + post-§21-sync hit), dig.fetchRange, and the mTLS peer range-stream (stream_range — the busiest node-to-node egress path, since multi-source downloaders fan byte-ranges across it).
• providers[].addresses inherit the DHT's IPv6-first candidate ordering (§5.2) — the throttle does not reorder them.

Blast radius (gitnexus impact analysis, §2.0)

Ran npx gitnexus@latest analyze on the repo, then gitnexus impact (CLI) on the touched symbols before editing:

• Node struct: exactly 5 direct construction sites, all in crates/dig-node-core/src/lib.rs (the from_env production constructor + 4 test constructors) — all updated with the new outgoing_throttle field.
• fetch_range_frame: 2 real call sites (lib.rs dispatch, peer.rs::NodeResponder::stream_range) — both updated.
• stream_range: a trait method (PeerRpcResponder) with independent per-type impls (StubResponder, integration-test responders) — editing the real NodeResponder impl does not affect the others.
• redirect_error_object: 1 existing caller — my new call sites are additive.
• Cross-crate check (grep across dig-node-service/dig-runtime): both only consume Node opaquely + the free functions handle_rpc/cache_cap_bytes/cache_used_bytes — none of the signatures touched here cross the crate boundary. Risk contained to dig-node-core (lib.rs, peer.rs, new bandwidth.rs).
• gitnexus detect-changes --scope compare --base-ref main post-commit reports "critical" risk / 71 affected flows — this is the Node struct's inherent connectivity (touching its private-field literal necessarily touches the 5 constructor sites and everything the two gap_fill/chain_watch end-to-end tests exercise), not a functional regression: the full workspace test suite (below) is green.

How verified

TDD throughout (failing test first, then implementation):

• bandwidth.rs: 8 new unit tests for the throttle's window/env-parsing logic (pure, no sleeps — *_at variants take the instant explicitly).
• lib.rs: 6 new integration tests over the real dig.getContent/dig.fetchRange dispatch — cap-exceeded+provider→redirect, no-provider fallback, no-P2P-engine fallback, under-cap non-interference, and hop-cap reuse (verified genuinely RED before wiring the dispatch, then GREEN after).
• peer.rs: 2 new tests over the real NodeResponder::stream_range (via tokio::io::duplex + serve_one_stream) — over-cap+provider→redirect frame, no-provider→streams anyway.
• cargo fmt --all -- --check: clean.
• cargo clippy --workspace --all-targets -- -D warnings: clean.
• cargo build --workspace: clean.
• cargo test -p dig-node-core (lib + tests/dht_integration.rs + tests/peer_network.rs + doctests): 209 + 8 + 2 passed, 0 failed.
• cargo test --workspace (dig-node-service, dig-runtime, dig-wallet): all green, no regressions.

Docs / SPEC

• SPEC.md: new §17 "Outgoing-bandwidth throttle and redirect-on-saturation" (config, accounting, serve-path integration, hop-cap reuse, graceful fallback); DIG_NODE_MAX_OUTGOING_BYTES_PER_SEC added to the §3.2 variables table; -32008 CONTENT_REDIRECT added to the §10 error catalogue (previously undocumented despite already being emitted by #165 redirect-on-miss).
• README.md: new env-var row.
• docs.dig.net peer-network page is not updated in this PR per the task's "ONLY dig-node; never another repo" scope constraint (and that repo currently has unrelated in-flight uncommitted changes on another branch) — flagging as a follow-up for whoever next touches docs.dig.net's peer-network/error-catalogue pages.

Version bump

crates/dig-node-core/Cargo.toml: 0.2.0 → 0.3.0 (minor — a new, backwards-compatible, opt-in capability: feat:, matching this repo's precedent, e.g. 0.1.0 → 0.2.0 for the dig.getManifest feature). The workspace root Cargo.toml has no [package]/[workspace.package] version field, so the ensure-version-increment gate is a no-op there per its own documented behavior.

Test plan (for reviewers)

• cargo test -p dig-node-core green (209 lib + 8 + 2 integration)
• cargo test --workspace green
• cargo fmt --all -- --check clean
• cargo clippy --workspace --all-targets -- -D warnings clean
• cargo build --workspace clean
• CI required checks (fmt / clippy / test+coverage / commitlint / version-increment) — confirming green after opening

Closes/references DIG-Network/dig_ecosystem#30.