Add native self-hosted instance connection to fluxer_desktop

Trimmed monorepo checkout (fluxer_desktop + packages/voice_engine_v2 +
tools/ci) with a "Connect to a Different Server" menu item and popout
that lets the desktop app switch to any self-hosted Fluxer instance,
plus fixes for well-known discovery on single-domain self-hosted
deployments and a false-positive ERR_ABORTED on same-origin client
redirects during the switch. Defaults to chat.fluxr.chat and uses an
isolated userData directory from the official build.

fluxer_desktop/native/screen-frame-bus/benches/baseline.json

@@ -0,0 +1,25 @@
+{
+	"measured_at": "85e057a273fd",
+	"host": "darwin-arm64-apple-silicon",
+	"regression_budget_percent": 5.0,
+	"criterion_args": {
+		"warm_up_time_sec": 2,
+		"measurement_time_sec": 5
+	},
+	"benches": {
+		"staging_pair_submit_map_cycle_1080p_rgba": {
+			"median_ns": 76243.0,
+			"low_ns": 72391.0,
+			"high_ns": 81080.0,
+			"budget_percent_override": 20.0,
+			"note": "1080p RGBA copy + map; observed cross-run sigma ~10% on a loaded dev host. Wider budget absorbs that without hiding real 1.3x slowdowns."
+		},
+		"frame_pool/acquire_release/cpu_1080p_rgba": {
+			"median_ns": 11.874,
+			"low_ns": 11.811,
+			"high_ns": 11.941,
+			"budget_percent_override": 10.0,
+			"note": "~12ns op; tiny absolute deltas easily exceed 5% percent. 10% trades sensitivity for noise floor."
+		}
+	}
+}

fluxer_desktop/native/screen-frame-bus/benches/frame_pool.rs

@@ -0,0 +1,26 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+use std::hint::black_box;
+
+use criterion::{Criterion, criterion_group, criterion_main};
+use fluxer_screen_frame_bus::frame_pool::{CpuFrameBuilder, FRAME_BYTES_MAX};
+
+fn bench_acquire_release_cpu_1080p_rgba(c: &mut Criterion) {
+    assert_eq!(FRAME_BYTES_MAX, 1920 * 1080 * 4);
+    let pool = CpuFrameBuilder::build_pool(FRAME_BYTES_MAX)
+        .expect("pool must allocate at construction time");
+    assert!(pool.capacity() > 0);
+
+    c.bench_function("frame_pool/acquire_release/cpu_1080p_rgba", |b| {
+        b.iter(|| {
+            let frame = pool
+                .try_acquire()
+                .expect("steady state single-threaded never starves");
+            black_box(frame.slot_index());
+        });
+    });
+    assert_eq!(pool.currently_in_flight(), 0);
+}
+
+criterion_group!(benches, bench_acquire_release_cpu_1080p_rgba);
+criterion_main!(benches);

fluxer_desktop/native/screen-frame-bus/benches/staging.rs

@@ -0,0 +1,50 @@
+// SPDX-License-Identifier: AGPL-3.0-or-later
+
+use criterion::{Criterion, black_box, criterion_group, criterion_main};
+use fluxer_screen_frame_bus::{CpuStagingBackend, StagingSurfacePair};
+
+const BENCH_WIDTH: usize = 1920;
+const BENCH_HEIGHT: usize = 1080;
+const BENCH_BYTES_PER_PIXEL: usize = 4;
+const BENCH_FRAME_LEN: usize = BENCH_WIDTH * BENCH_HEIGHT * BENCH_BYTES_PER_PIXEL;
+
+fn build_pair() -> StagingSurfacePair<CpuStagingBackend> {
+    let backends = [
+        CpuStagingBackend::new(BENCH_FRAME_LEN),
+        CpuStagingBackend::new(BENCH_FRAME_LEN),
+    ];
+    assert_eq!(backends[0].len(), BENCH_FRAME_LEN);
+    assert_eq!(backends[1].len(), BENCH_FRAME_LEN);
+    StagingSurfacePair::new(backends)
+}
+
+fn submit_map_cycle(c: &mut Criterion) {
+    let mut pair = build_pair();
+    pair.submit(0, |buf| fill_pattern(buf, 0))
+        .expect("warm submit zero");
+
+    let mut sequence: u64 = 1;
+    c.bench_function("staging_pair_submit_map_cycle_1080p_rgba", |b| {
+        b.iter(|| {
+            pair.submit(sequence, |buf| fill_pattern(buf, sequence as u8))
+                .expect("submit ok in steady state");
+            let mapped_sequence = sequence - 1;
+            let first_byte = pair
+                .try_map(mapped_sequence, |buf| buf[0])
+                .expect("cpu backend is always ready");
+            black_box(first_byte);
+            sequence = sequence.wrapping_add(1);
+        });
+    });
+}
+
+fn fill_pattern(buf: &mut [u8], seed: u8) {
+    assert_eq!(buf.len(), BENCH_FRAME_LEN, "bench backend has fixed size");
+    let chunk = seed.wrapping_add(1);
+    for byte in buf.iter_mut() {
+        *byte = chunk;
+    }
+}
+
+criterion_group!(benches, submit_map_cycle);
+criterion_main!(benches);