SASEUL GPU Miner 3.5.4-beta -- NVIDIA Linux x86-64 -- no devfee

Experimental opt-in release. Stable 3.5.3 remains unchanged.
Extract: tar xzf saseul-miner-hive-3.5.4-beta.tar.gz
Run: cd saseul-miner-hive && ./saseul-miner --wallet YOUR_SASEUL_WALLET --id YOUR_WORKER
HiveOS Custom Miner installation URL:
https://saseulpool.com/files/saseul-miner-hive-3.5.4-beta.tar.gz
Use your own wallet and worker, and existing pool settings. No auto-update.
Keep original-miner and f11-miner BESIDE saseul-miner; all three are required.
Ubuntu 22.04-compatible x86-64 runtime (glibc >=2.34, libstdc++ GLIBCXX_3.4.30,
OpenSSL 3 libcrypto.so.3), compatible NVIDIA driver; not ARM64/Jetson/Windows.
Older HiveOS images may need a compatible OS runtime; no driver changes included.
Hive integration also uses Python 3 (safe extra-argument parsing), jq and bc.
Beta Hive helpers recognize either backend, preserve the beta version label,
quote generated config values, and never kill other miners by command pattern.

Dispatcher selects f11 on measured CUDA capabilities 7.5/8.0/8.6/8.9/12.0.
Other or mixed fast/fallback GPU groups use the original 3.5.3-source backend.
One backend process handles the selected GPUs. Per-GPU mixed-rig optimized
routing is not implemented. CUDA_VISIBLE_DEVICES and --gpus are respected.
Latehex is default; corrected nonce is 24+16=40 hex characters. CPU proof
verification, full targets, NATS and idle fallback are retained.

Measured independent warmed ABBAABBA, unchanged settings per card:
RTX 2070 SUPER: 4.4803375 -> 5.263369 GH/s (+17.4771%)
A100 SXM4 40GB: 9.554632 -> 12.1552375 GH/s (+27.2183%)
RTX 3070 Ti: 4.428188 -> 5.708661 GH/s (+28.9164%)
RTX 4090: 19.2682795 -> 21.036632 GH/s (+9.1775%)
RTX 5090: 26.044506 -> 27.326936 GH/s (+4.9240%)
These are offline hashing rates, NOT income or guaranteed gains for all models.
Full-digest/OpenSSL and target-boundary tests passed; short live tests produced
server-accepted PPLNS shares, observed hw_err=0. No long soak certification or
new canonical block/income claim. x1.5--2 optimization target not achieved.

Tesla M40/GTX1070: original backend only (f11 regresses); their baseline live
tests stopped at thermal guard and were not completed soak tests.
17 CUDA native targets compile, but only the named tested models have runtime
evidence. Do not mistake compilation for testing every card. AMD unsupported
in this release; its separate OpenCL work is not included.
Keep stable package available for rollback; do not overwrite stable aliases.
