bench: make the CUDA embedding path discoverable when it is unavailable

The GPU path worked but was effectively hidden. cudarc's build script shells out
to `nvcc`, which ships in /usr/local/cuda/bin — a directory the reference host
had installed but never exported to the login shell, so `--features
embeddings-cuda` failed with a bare "`nvcc --version` failed" panic from a
dependency's build script, and the runtime fallback then reported only
"Embedder: CPU (...)" before spending hours on work a GPU does in minutes.

Two changes, both about making the failure legible rather than changing what the
code does:

  - The CPU fallback now says why it fell back and what that costs, with the
    concrete fix. A run that silently takes two orders of magnitude longer reads
    as a hang, not as a configuration choice.
  - BENCHMARKS.md states the build-time nvcc requirement, where the toolkit
    actually installs, and that a shell file read non-interactively is the place
    to export it — `~/.zshenv` rather than `~/.zshrc`, because build scripts do
    not run in an interactive shell.

Host-side, the reference machine's CUDA exports lived in ~/.bashrc below its
non-interactive guard while the login shell is zsh, so they never applied to
anything. Moved to ~/.zshenv with duplicate-prepend guards; `nvcc --version`
and `cargo build --features embeddings-cuda` now both work over a plain
non-interactive ssh with no manual export.

BENCHMARKS.md

@@ -300,8 +300,17 @@ matching, and MiniLM at 384 dimensions is a small embedding model.
 
 > **Run:** `cargo run --release --bin longmemeval_bench --features embeddings -- \
 >   benchmarks/longmemeval/longmemeval_s_cleaned.json --embeddings weights/all-minilm-l6-v2`
-> Add `--features embeddings-cuda` (and put `nvcc` on `PATH`) for the GPU path.
+> For the GPU path use `--features embeddings-cuda`. That requires `nvcc` on
-> Weights: `huggingface.co/sentence-transformers/all-MiniLM-L6-v2`.
+> `PATH` at *build* time — cudarc's build script shells out to it. The toolkit
+> installs to `/usr/local/cuda/bin`, which many distributions do not export;
+> check with `nvcc --version` and, if it is missing, add it somewhere every
+> shell reads (for zsh that is `~/.zshenv`, not `~/.zshrc`, since build tooling
+> runs non-interactively). The device is selected at runtime with a CPU
+> fallback, so a machine without CUDA still produces correct numbers — just far
+> more slowly, and the bench says so on startup.
+>
+> Weights: `huggingface.co/sentence-transformers/all-MiniLM-L6-v2` — place
+> `model.safetensors` and `tokenizer.json` in the `--embeddings` directory.
 
 ### Oracle variant — `longmemeval_oracle`, n=500 (easier corpus, kept for continuity)
 

crates/clawhdf5-bench/src/bin/longmemeval_bench/embedder.rs

@@ -43,7 +43,13 @@ impl Embedder {
                 d
             }
             Err(e) => {
-                eprintln!("Embedder: CPU ({e})");
+                // Loud, because the CPU path is correct but ~100x slower: the
+                // full longmemeval_s haystack is minutes on a GPU and most of a
+                // day on 8 cores. Silently falling back looks like a hang.
+                eprintln!("Embedder: CPU — CUDA unavailable ({e})");
+                eprintln!(
+                    "  WARNING: CPU embedding is roughly two orders of magnitude slower.\n                       Expect minutes for longmemeval_oracle and many hours for the full\n                       longmemeval_s haystack. For the GPU path, rebuild with\n                       `--features embeddings-cuda` and make sure `nvcc` is on PATH\n                       (it ships in /usr/local/cuda/bin, which is often not exported)."
+                );
                 Device::Cpu
             }
         };