net.http: enable HTTP/2 by default for https requests

[quaesitor-scientiam]

What

Flips the enable_http2 default (on Request and FetchConfig) from false
to true, so http.fetch / http.get to an https:// URL negotiate HTTP/2
when the server offers it, and fall back to HTTP/1.1 otherwise. To force
HTTP/1.1, set enable_http2: false.

resp := http.get('https://example.com/')        // HTTP/2 if the server supports it
resp.version() // .v2_0 on most modern servers, .v1_1 otherwise

forced := http.fetch(url: 'https://example.com/', enable_http2: false)! // always HTTP/1.1

This is the opt-out counterpart to the opt-in support added in #27362, now that
the HTTP/2 client also handles redirects, streaming callbacks (#27369), and flow
control, and has been exercised against real servers.

Scope

The change is three lines (the two field defaults + the doc) plus a test update.

File	Change
request.v, http.v	enable_http2 bool → enable_http2 bool = true, with updated docs
h2_client_test.v	the gated real-server test now asserts a default get() negotiates HTTP/2 and that enable_http2: false forces HTTP/1.1

Behaviour / compatibility

• Unaffected: plain http:// (no ALPN, never HTTP/2), and the Windows
  SChannel backend, which has no ALPN yet and stays on HTTP/1.1 (tracked in
  net.http: add ALPN (and HTTP/2) support to the Windows SChannel TLS backend #27383).
• No latency regression: the client sends the connection preface, its
  SETTINGS, and the request together without waiting for the server's SETTINGS,
  so there's no added round trip; HPACK also shrinks request headers.
• Transparent fallback: advertising h2, http/1.1 is safe — servers that
  don't understand ALPN simply ignore it and the client uses HTTP/1.1.

Honest tradeoff for the maintainer to weigh

Each fetch is still one connection / one request (no pooling or
multiplexing yet), so HTTP/2's headline win — multiplexing many requests over
one connection — doesn't yet materialize. This PR is purely about defaulting to
the newer protocol; pooling/multiplexing is a planned follow-up. If you'd rather
keep HTTP/2 opt-in until pooling lands, this PR can simply wait — the opt-in
path (#27362) already works.

Tests

• Full vlib/net/http suite green (mbedtls; 1 Windows-only skip).
• The existing real-network HTTPS test suite (http_httpbin_test.v:
  GET/POST/PUT/PATCH/DELETE/HEAD against a live server) passes through the new
  default HTTP/2 path unchanged.
• -d network test confirms a default get('https://www.google.com/')
  returns HTTP/2.0, and enable_http2: false returns HTTP/1.1.
• -W -cstrict -cc clang clean.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.

[quaesitor-scientiam]

• Codex Review Findings

• Medium: The updated -d network test will fail on default Windows builds. get('https://www.google.com/') at vlib/net/http/
  h2_client_test.v:89 now expects .v2_0, but Windows still routes HTTPS through SChannel at vlib/net/http/backend.c.v:10. That backend
  builds a plain HTTP/1.1 request and has no ALPN/HTTP/2 path at vlib/net/http/backend_vschannel_windows.c.v:21. The test needs a Windows/
  no_vschannel gate or the SChannel path needs HTTP/2 support.

• Low: The comment in vlib/net/http/backend.c.v:24 is now stale: it says ALPN is advertised “only when HTTP/2 is requested” so existing
  callers see no wire change. With enable_http2 defaulting to true, ordinary get()/fetch() calls now advertise ALPN on non-SChannel HTTPS.

Tests Run

• .\vnew.exe test vlib\net\http\h2_client_test.v passed; the -d network block was skipped.
• .\vnew.exe fmt -verify vlib\net\http\http.v vlib\net\http\request.v vlib\net\http\h2_client_test.v passed.
• git diff --check master...HEAD passed.

[quaesitor-scientiam]

Both addressed in fa3d8c1:

Medium (Windows network test): test_http2_fetch_real_server now skips on the default Windows/SChannel configuration ($if windows && !no_vschannel ?), since SChannel has no ALPN yet (#27383) so a default get() there stays on HTTP/1.1. Still covered with -d no_vschannel (mbedtls client).

Low (stale comment): Updated the comment in backend.c.v — it no longer claims "no change on the wire"; it now notes that with enable_http2 defaulting to true, ordinary get()/fetch() advertise ALPN and use HTTP/2 when the server selects it, with opt-out via enable_http2: false.

Verified: full vlib/net/http suite green; -d network test passes (non-Windows); fmt -verify clean.