Adds a `throughput` benchmark to every target and closes two small
test-coverage documentation/QA gaps. One PR, no merge of binding code beyond
the additive benchmarks and one C test.
## 1. Per-binding throughput benchmarks (all 9 targets)
Each benchmark feeds a deterministic synthetic OHLCV series through three
indicators chosen by **FFI call-signature archetype** (not algorithm — the same
Rust core runs underneath all bindings):
- `SMA(20)` — 1-in → 1-out (baseline boundary cost)
- `ATR(14)` — multi-in → 1-out (input marshalling)
- `MACD(12,26,9)` — 1-in → multi-out (output marshalling)
Streaming is timed for all three; batch for the single-output SMA and ATR
(median of 3 runs, after a warmup pass).
New: Python (PyO3), WASM, C (CMake), C# (Stopwatch), Go, Java (FFM), R, and the
Rust core baseline (`examples/rust/.../throughput.rs`, **no FFI** — the ceiling
the bindings are measured against and the value their batch paths converge
towards). Node already had `throughput.js`.
**Not a speed claim:** there is no comparable streaming TA library for C, C#,
Go, Java, R or WASM to compare against, so these are raw per-binding throughput
numbers documenting each language's FFI overhead — see BENCHMARKS.md §3. The
"Wickra is fast" claim still lives in §1/§2 (Rust core + the Python/Rust
cross-library runs).
## 2. README `## Testing`: C# and C bullets
The section listed every layer except C# and C, even though both have suites.
Adds the two missing bullets.
## 3. C archetype ctest
`examples/c/archetypes.c` drives one indicator per FFI archetype through the
real C boundary (scalar + batch==streaming, multi-output, bars, profile, array
input) plus reset, invalid-parameter and NULL-safety — the C counterpart of the
Go/R/Java archetype suites. Runs on three OSes via the existing CMake/ctest.
## Notes
- Benchmarks are not CI-gated (manual-run scripts, like the existing
`throughput.js`); no `ci.yml`/`release.yml` changes.
- Docs: BENCHMARKS.md §3, a `## Benchmark` section in every binding README, a
CHANGELOG entry.
- Verified locally by running: Rust, Python, C, C#, Go, Java (real numbers); the
C archetype ctest with `-Wall -Wextra -Wpedantic -Werror`. WASM and R are
API-correct and syntax-checked but need their own toolchains to run.
The published Python benchmark tables (README/BENCHMARKS.md) were a stale, incoherent run. Re-measured locally with the current build (wickra 0.6.5, post batch fast-paths) via `compare_libraries.py` on the same 9950X.
- **Streaming vs talipp:** 11-56x (was 9-58x).
- **Batch:** real per-indicator numbers; MACD and ATR were notably off in the old table.
- **Prose:** Wickra beats TA-Lib on RSI and ATR (no longer MACD, which now trails 130 vs 111 us).
Rust tables unchanged. Numbers are a single coherent run; absolute us still depend on machine state (caveat already in the doc).
## Summary
- Dedicated batch fast paths for **EMA, RSI, Bollinger, MACD and ATR** (used by the Python bindings): one allocation filled in a single pass, warmup encoded as `NaN`, no per-element `Option` or input re-validation. Each is **bit-for-bit equal** to replaying `update` — SMA/Bollinger keep the drift-reseed cadence, the EMA-family keep the seed division and `mul_add` recurrences. Adds the `BatchNanExt` extension trait.
- **Cross-library benchmark refresh**: `compare_libraries.py` reports the median across timing rounds (`--rounds` / `--streaming-rounds`), gains `--skip-batch` / `--skip-streaming`, and runs every peer through the streaming arena (recompute for batch-only libraries). `wickra-bench` drives the batch fast paths against `kand`.
- **README** benchmark section reordered streaming-first (the order-of-magnitude result), with measured TA-Lib/tulipy/pandas-ta numbers in place of the CI-only placeholders.
## Impact
- Python batch ~2× faster on EMA/RSI/MACD/ATR; streaming path unchanged.
- The `batch == streaming` equivalence stays bit-exact.
## Verification
- `cargo fmt` · `cargo clippy --workspace --all-targets --all-features -- -D warnings` (clean)
- `cargo test --workspace --all-features` — 3782 unit + 420 doc tests pass
- Python `pytest` — streaming-vs-batch, known-values, input-validation, smoke pass
## Notes
- Node/WASM bindings keep their existing batch; the fast paths are Python-only for now.
