fix(http): get-stream yields lines on newline, not buffer-fill (ILO-489)

[danieljohnmorris]

Summary

get-stream / pst-stream (ILO-448) yield lines lazily, but the native backend wrapped minreq's ResponseLazy in BufReader::lines(), which blocks on a full ~8 KiB read-buffer fill before surfacing any line. So a slow SSE upstream that flushes one short event then idles had its lines batched until the buffer filled or the connection closed. Functionally correct (all lines arrive in order) but latency was buffer-bound, not event-bound.

Motivation: this directly unblocks ILO-482's end-to-end test, which proxies an upstream via get-stream to prove httpd lazy-body streaming and was flaky (~1 in 3) because get-stream sometimes hadn't yielded the first line when the test read. Per-newline yield makes that test deterministic.

Repro before/after

A server flushes first\n, idles 300ms, then sends second\n without closing.

• Before: the client doesn't surface first until second (or close) arrives - line 1 lands ~200us after line 2 hits the wire (batched).
• After: the client surfaces first the instant its \n arrives, well before line 2 is sent.

What changed (per commit)

• fix(http): stream get-stream lines on newline, not buffer-fill - replace the BufReader::lines() wrapper with a LazyLineSplitter that consumes ResponseLazy's byte iterator incrementally, emitting each line the moment a \n is seen. Strips a trailing \r for chunked/CRLF, yields any partial line at EOF, maps minreq::Error::IoError back to the io::Error the existing ILO-R009 http-stream read error path expects. Covers all four builtins (get-stream, get-stream-h, pst-stream, pst-stream-h).
• test(http): prove get-stream yields per-newline, not per-buffer - slow-drip server reports the instant line 2 hits the wire; the test asserts line 1 arrived before then with margin. Verified to fail on the old buffered code and pass on the fix. Plus a trailing newline-less EOF line test.
• docs: note get-stream per-newline yield (ILO-489) - CHANGELOG Unreleased entry + skill doc latency note.

Test plan

• cargo fmt
• cargo clippy --release --all-targets --features cranelift -- -D warnings clean
• cargo test --release --features cranelift green (406 lib + all integration; only the two tolerated baseline doctests in src/pkg.rs / src/verify.rs fail, as on main)
• new prompt-yield test fails on old code, passes on fix
• existing ILO-448 streaming tests still pass (no regression)
• rust-review subagent: approve, no changes required

Follow-ups

None.

[codecov]

Codecov Report

❌ Patch coverage is 80.55556% with 7 lines in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/interpreter/http_wasm.rs	80.55%	7 Missing ⚠️

📢 Thoughts on this report? Let us know!