feat: adjustable radius API (#18)

* fix: add radius parameter to FrameMonitor and threshold to FrameCoverage

  Previously FrameMonitor used hardcoded RADIUS from env var, inconsistent
  with BKFrameMonitor and RustBKFrameMonitor which accept radius in their
  constructors. This change makes all monitors consistent.

  - Add FrameMonitor(radius=) constructor parameter
  - Add FrameCoverage(threshold=) constructor parameter
  - Add threshold parameter to get_frame_cov()
  - Create docs/api.md with comprehensive API reference
  - Update docs/frame_cov.md configuration section

* ignore fuzz link

* radius API analysis

* feat: change default RADIUS from 5 to 10

  Based on empirical analysis of Zelda fuzzing dataset:
  - radius=5: linear growth, no convergence (growth rate 0.55)
  - radius=10: converging behavior (growth rate 0.36)

  The new default provides better convergence characteristics
  for games with high visual entropy while still distinguishing
  meaningfully different game states.

  See radius_analysis_report.md for full experimental results.

* bump version

Author: EYH0602

.gitignore

@@ -20,3 +20,5 @@ prof/
 # Rust build artifacts
 target/
 Cargo.lock
+
+fuzz

AGENTS.md

@@ -53,7 +53,9 @@ Future metrics (e.g., audio coverage, state-graph coverage) will follow the same
 │   ├── videos/                  # Small sample MP4s for integration tests
 │   └── smb/                     # Super Smash Bros recordings for stress tests
 ├── docs/
-│   └── design.md                # Architecture and design documentation
+│   ├── api.md                   # API reference documentation
+│   ├── frame_cov.md             # Architecture and design documentation
+│   └── tuning.md                # Radius/threshold tuning guide with empirical results
 ├── rustfmt.toml                 # Rust formatting config
 ├── .pre-commit-config.yaml      # Pre-commit hooks (Python + Rust)
 ├── .github/workflows/           # CI: pytest, mypy, ruff, pylint, rust (fmt/clippy/test)
@@ -74,7 +76,9 @@ Run Rust linting: `cargo clippy --all-targets --all-features -- -D warnings`
 
 ## Design
 
-See [docs/design.md](docs/design.md) for the coverage framework architecture, frame coverage pipeline, BK-tree optimization, and loading strategies.
+See [docs/frame_cov.md](docs/frame_cov.md) for the coverage framework architecture, frame coverage pipeline, BK-tree optimization, and loading strategies.
+
+See [docs/api.md](docs/api.md) for detailed API reference documentation including configuration options.
 
 ## Key Modules (quick reference)
 
@@ -88,7 +92,7 @@ See [docs/design.md](docs/design.md) for the coverage framework architecture, fr
 | `writer.py` | `write_mp4()`, `write_mp4_cv2()` |
 | `stitch.py` | `stitch_images()` (panorama via AffineStitcher) |
 | `generator.py` | Hypothesis strategies: `frames()`, `frames_lists` |
-| `env.py` | `RADIUS` env var (Hamming distance threshold, default `5`) |
+| `env.py` | `RADIUS` env var (default Hamming distance threshold, `10`); use constructor params to override |
 
 ## Environment and Dependencies
 
@@ -142,7 +146,7 @@ Some tests require assets in `assets/videos/` or `assets/smb/` and will skip if
 
 ### Environment variables for tests
 
-- `RADIUS` — Hamming distance threshold (default `5`).
+- `RADIUS` — Default Hamming distance threshold (default `10`). Prefer passing `radius=` to monitor constructors.
 - `N_MAX` — Maximum number of recordings to process in monotonicity tests (default `100`).
 
 ## Benchmarks