test(driver): branch coverage >=80% on src/driver/{http2,undici}

[productdevbook]

Summary

Drives previously-uncovered branches in src/driver/http2.ts and src/driver/undici.ts so each file lands at >=80% branch coverage and the src/driver aggregate clears the same bar. No source changes — tests only.

Coverage (before -> after, branches)

File	Before	After
src/driver/http2.ts	60.37%	81.13%
src/driver/undici.ts	70.00%	80.00%
src/driver/ overall	64.38%	80.82%

Measured via pnpm vitest run --coverage --coverage.include='src/driver/**'.

Scenarios added

http2.ts:

• host header stripped before sending (server sees only :authority)
• multi-value response header append (set-cookie array path)
• server-initiated RST_STREAM rejects via the error listener
• pre-aborted signal rejects synchronously (queueMicrotask path)
• cached session reuse across sequential requests (bumpIdle on hit)
• recovery when the cached session was destroyed between calls
• idle-timer expiry closes the session, next call opens a fresh one
• dispose() is idempotent and a no-op before any request

undici.ts:

• POST forwards body bytes and sets duplex: 'half'
• bodyless GET omits both body and duplex
• request.signal forwards through to the dispatcher
• dispatcher object identity preserved (same reference, not cloned)
• POST/PUT/DELETE method propagation
• request.redirect propagation
• ensureFetch caches the import across concurrent requests
• real-undici end-to-end POST (covers dynamic-import branch)
• real-undici non-2xx surfaces as HTTPError
• error path when undici exposes no fetch

Test plan

• pnpm vitest run test/driver-undici.test.ts test/driver-http2.test.ts — 30/30 pass
• pnpm vitest run — full suite 932/932 pass (twice in a row)
• pnpm lint — no new errors introduced
• pnpm fmt — formatted clean
• http2 sockets destroyed manually on teardown (per 1ba17a7 pattern)

🤖 Generated with Claude Code

[github-actions]

Bench results

✅ No misina row regressed >5% vs base.

benchmark	base avg	head avg	Δ
steady-state GET (200 OK, JSON parse) :: axios	308.45 µs	311.89 µs	· +1.1%
steady-state GET (200 OK, JSON parse) :: misina	303.97 µs	306.67 µs	· +0.9%
steady-state GET (200 OK, JSON parse) :: ky	291.92 µs	291.61 µs	· -0.1%
retry on 503 → 200 :: misina — retry 1× then 200	1.71 ms	1.71 ms	· -0.4%
POST JSON body :: native fetch	367.46 µs	365.60 µs	· -0.5%
steady-state GET (200 OK, JSON parse) :: ofetch	270.16 µs	268.52 µs	· -0.6%
POST JSON body :: ofetch	339.42 µs	336.36 µs	· -0.9%
POST JSON body :: ky	463.77 µs	457.06 µs	· -1.4%
POST JSON body :: axios	326.03 µs	319.16 µs	· -2.1%
createMisina cold start :: createMisina({ use: [bearer, tracing, breaker] })	1.85 µs	1.80 µs	· -3.2%
steady-state GET (200 OK, JSON parse) :: native fetch	341.89 µs	326.14 µs	· -4.6%
plugin overhead (no plugins / 1 hook plugin / 3 plugins inc. wrapping) :: misina — bearer + tracing + breaker	321.71 µs	305.23 µs	🟢 -5.1%
plugin overhead (no plugins / 1 hook plugin / 3 plugins inc. wrapping) :: misina — no plugins	291.19 µs	272.64 µs	🟢 -6.4%
hooks overhead (no hooks vs 5 hooks) :: misina — 5 hooks	303.57 µs	283.50 µs	🟢 -6.6%
plugin overhead (no plugins / 1 hook plugin / 3 plugins inc. wrapping) :: misina — bearer	295.73 µs	273.59 µs	🟢 -7.5%
createMisina cold start :: createMisina({ use: [bearer] })	553.7 ns	510.7 ns	🟢 -7.8%
hooks overhead (no hooks vs 5 hooks) :: misina — no hooks	312.07 µs	281.67 µs	🟢 -9.7%
POST JSON body :: misina	634.11 µs	503.63 µs	🟢 -20.6%
createMisina cold start :: createMisina()	1.20 µs	535.4 ns	🟢 -55.5%

_{Local Node HTTP server, Apple-class GitHub runner. Numbers fluctuate ±2-3% from runner heat alone — only sustained >5% deltas are signal.}

[productdevbook]

Driver test coverage looks solid: real http2 server (with proper socket cleanup destroying every accepted socket on teardown — necessary because Http2Server.close lacks closeAllConnections), real http listener for undici. Branch coverage hits the targets (http2 81.13%, undici 80%, driver/ 80.82%). Tests exercise real async paths: RST_STREAM via stream.close(0x02), abort via DOMException AbortError, ensureFetch concurrent-import branch, dispatcher identity propagation, multi-value set-cookie via Headers.append. Disposal is tested for idempotency and pre-request no-op. Caught and fixed one issue: a test referenced followRedirects (not a real MisinaOptions field) — replaced with redirect: 'follow'. Full suite 932/932 green after fix.