chore(release): v1.9.21 — skip H2 for full-tunnel batches (#1040)

Bumps Cargo.toml v1.9.20 → v1.9.21 and ships the changelog. Headline:
PR #1040 (@yyoyoian-pixel) skips H2 multiplexing for tunnel_batch_request_to
(Full-mode batch path). Tunnel batches already coalesce N ops into one
HTTP request, so H2 stream multiplexing has nothing to multiplex there;
the H2 path was actively hurting via long-poll stalls + silent batch
drops + pool starvation. H2 remains active for relay mode (apps_script),
where r0ar's #962 A/B data confirmed it's strictly better. A/B on Pixel
6 Pro: 0/30 vs 8-10/30 long-poll stalls.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/changelog/v1.9.21.md

@@ -0,0 +1,31 @@
+<!-- see docs/changelog/v1.1.0.md for the file format: Persian, then `---`, then English. -->
+• **Perf: skip H2 برای Full-tunnel batch requests** ([PR #1040](https://github.com/therealaleph/MasterHttpRelayVPN-RUST/pull/1040) by @yyoyoian-pixel). Full mode tunnel batches قبلاً N op رو در یک HTTP request coalesce می‌کنند — H2 stream multiplexing چیزی برای multiplex کردن نداره. H2 try/fallback path در این مسیر خاص سه regression از v1.9.14 معرفی کرد: (1) long-poll stallها در ۱۶-۱۷s به جای 10s timeout روی H1 — هر poll ~۶۰٪ بیشتر slot Apps Script رو نگه می‌داشت، (2) silent batch drops via \`RequestSent::Maybe\` بدون retry، (3) pool starvation از \`POOL_MIN_H2_FALLBACK = 2\` که از 8 → 2 trim می‌کرد. H2 multiplexing برای **relay mode** (apps_script) فعال می‌مونه — اونجا واقعاً به‌درد می‌خوره (r0ar در #962 confirmed). A/B روی Pixel 6 Pro: **0/30 vs 8-10/30** long-poll stalls. ۲۰۹ lib test still pass. v1.9.14 tunnel performance بازگشت + همه v1.9.15+ improvements حفظ شد (relay mode h2، zero-copy mux، block DoH/QUIC، PR #1029 warm-race fix).
+---
+• **Perf: skip H2 for Full-tunnel batch requests** ([PR #1040](https://github.com/therealaleph/MasterHttpRelayVPN-RUST/pull/1040) by @yyoyoian-pixel). Tunnel batches already coalesce N ops into one HTTP request — H2 stream multiplexing has nothing to multiplex on this code path. The H2 try/fallback block introduced three regressions vs v1.9.14:
+
+1. **Long-poll stalls**: idle polls completed at 16-17s (`LONGPOLL_DEADLINE` + network latency) instead of timing out at 10s on H1. Each poll held an Apps Script execution slot ~60% longer.
+2. **Silent batch drops**: `RequestSent::Maybe` failures dropped the entire batch with no retry — a failure mode H1 doesn't have.
+3. **Pool starvation**: `POOL_MIN_H2_FALLBACK = 2` trimmed the H1 pool from 8 → 2 once H2 connected, but tunnel batches still used H1 and needed the full pool.
+
+H2 multiplexing **stays active for relay mode** (non-full) where each browser request is a separate HTTP call that genuinely benefits from stream multiplexing — r0ar's controlled A/B test in #962 confirmed h2 is strictly better than `force_http1: true` for apps_script-mode users, and that path is unchanged here.
+
+## Changes (`domain_fronter.rs`-only, -54/-12 lines, +12 net)
+
+- `tunnel_batch_request_to`: remove H2 try/fallback/NonRetryable block, go straight to H1 pool `acquire()`.
+- `run_pool_refill`: always maintain `POOL_MIN = 8`. Remove the `POOL_MIN_H2_FALLBACK = 2` trim.
+
+## A/B results (Pixel 6 Pro, 30 batch samples each)
+
+| Metric | H2 (stock v1.9.20) | H1 (this release) | v1.9.14 (baseline) |
+|---|---|---|---|
+| 16-17s batches | **8-10/30** | **0/30** | **0/30** |
+| 10s timeouts | 0 | 4/30 | 5/30 |
+| Active RTTs | 1.4-2.4s | 1.3-2.2s | 1.4-2.3s |
+
+Restores v1.9.14 tunnel performance while keeping all v1.9.15+ improvements (H2 for relay, zero-copy mux from PR #881, block DoH/QUIC defaults from v1.9.13/14, PR #1029's warm-race fix from v1.9.20).
+
+## Interaction with v1.9.20 (PR #1029)
+
+PR #1029 added `H2Cell.dead: Arc<AtomicBool>` for synchronous dead-cell detection. With this release removing the H2 path for tunnel batches, the dead-cell flag scopes to relay mode only — that's intentional (the flag was protecting the relay path in practice). No regression.
+
+209 lib tests still pass (no test changes — the affected paths are exercised by integration probes which the PR reporter ran on Pixel 6 Pro).