fix(dev): make the documented checks pass on a fresh clone

[RickyYii]

CONTRIBUTING asks contributors to run cargo test, cargo fmt -- --check and cargo clippy --all-targets -- -D warnings, and none of the three passes on a clean checkout today. cargo test never reaches a test on Linux or macOS: the Tauri build script aborts when a declared bundle resource path is missing, and src-tauri/tauri.linux.conf.json and src-tauri/tauri.macos.conf.json declare engines/linux/brush/ and engines/macos/arm64/, which are gitignored and therefore absent until setup:engines downloads binaries into them, so a fresh clone fails with resource path `../engines/linux/brush` doesn't exist before compiling anything; CI never sees this because it installs the engines first, so it only bites new contributors. This PR tracks a README in each of those two directories, following the convention already used for engines/ffmpeg, engines/colmap and engines/brush, so the directories survive a clone while the binaries stay out of Git. Once the build runs, two tests fail for reasons unrelated to the code under test: linux_root_is_flat_and_discovery_can_fall_back_to_path asserts discovered.ffmpeg.is_file(), which requires FFmpeg on PATH and is not stated as a prerequisite anywhere, so it now asserts the resolver contract instead — an explicit OOOSPLAT_FFMPEG override wins, then PATH, and otherwise the managed path is kept so engine health can report the exact file it expected — covering the same behaviour on a host with or without FFmpeg; and cancellation_terminates_descendant_processes asserts that /proc/{pid} no longer exists, but a killed descendant is reparented to PID 1 and PID 1 does not reap in every container, so the process correctly stops running yet lingers as a zombie with its /proc entry intact and the assertion fails, which is what happens in a plain container today. It now asserts termination directly: on Linux the entry is gone or the state is Z, and on other Unix targets signal 0 reports no such process. Worth flagging that the old assertion was also vacuous on macOS, where /proc never exists and exists() is always false, so that test asserted nothing there; the new one does assert, and while the reasoning holds — the orphaned sleep is reparented to launchd and reaped promptly, well inside the bounded three-second poll that replaces the previous fixed 50ms sleep — I have no macOS machine to confirm it on, so please do watch the macOS workflow on this PR. Finally, cargo clippy --all-targets -- -D warnings fails on current stable Rust at reserve_gaussian_video_export in src-tauri/src/commands/mod.rs with clippy::nonminimal_bool; the workflows install Rust with an unpinned rustup update stable, so this turns the Ubuntu and macOS workflows red as runners pick the toolchain up, and it is fixed here because otherwise this PR could not show a green run either. Verified on Ubuntu 24.04 x86_64 with Rust 1.94.1: cargo test 68 passed / 0 failed (66 passed / 2 failed before), cargo fmt -- --check clean, cargo clippy --all-targets -- -D warnings clean, npm test 48 passed, tsc -b clean. The fresh-clone claim was checked end to end by cloning the branch into an empty directory and running cargo test with no mkdir and no setup:engines: 68 passed. No pipeline stage was exercised — this change touches only test assertions, two tracked placeholder READMEs, a .gitignore negation and one boolean expression, and does not touch FFmpeg, COLMAP, Brush or PLY output.