chore(bench): restructure into focused targets, add runner, and tune slow cases (#10)

- Upgrade Criterion 0.5 → 0.8.2
- Split 4 broad bench files into 14 focused targets
- Add Rust runner binary with per-target start/done logging
- Add wrap_render benchmarks (all 4 WrapMode variants × 2 sizes)
- Add undo_redo benchmarks (undo/redo × 3 history depths)
- Apply benchmark_group + SamplingMode::Flat + reduced sample_size on slow insert cases to cut total run time from ~28 min to ~12 min
- Rewrite bench/README.md to document runner, targets, and workflow

bench/README.md

@@ -1,34 +1,74 @@
 Benchmarks for tui-textarea using [Criterion.rs][criterion].
 
-To run all benchmarks:
+## Running all benchmarks (Rust runner — recommended)
+
+The Rust runner executes every benchmark target sequentially and prints
+per-target start time, exit status, and wall duration.
 
 ```sh
-cargo bench --benches
+cargo run -p tui-textarea-bench --bin runner
 ```
 
-To run specific benchmark suite:
+Example output:
+
+```
+[1/12] START insert_append at 2026-02-18T12:00:00Z
+[1/12] DONE  insert_append status=ok duration=42.31s
+[2/12] START insert_random at 2026-02-18T12:00:42Z
+...
+All 14 benchmark targets completed successfully.
+```
+
+## Running a single benchmark target
 
 ```sh
-cargo bench --bench insert
+cargo bench -p tui-textarea-bench --bench insert_append
 ```
 
-To filter benchmarks:
+Available targets:
+
+| Target             | What it measures                          |
+|--------------------|-------------------------------------------|
+| `insert_append`    | Appending lorem lines sequentially        |
+| `insert_random`    | Inserting lorem lines at random positions |
+| `insert_long`      | Typing a long line without newlines       |
+| `search_forward`   | Forward regex search                      |
+| `search_backward`  | Backward regex search                     |
+| `cursor_char`      | Character-level cursor movement           |
+| `cursor_word`      | Word-level cursor movement                |
+| `cursor_paragraph` | Paragraph-level cursor movement           |
+| `cursor_edge`      | Head/end and top/bottom jumps             |
+| `delete_char`      | Deleting one character at a time          |
+| `delete_word`      | Deleting one word at a time               |
+| `delete_line`      | Deleting to line head                     |
+| `wrap_render`      | Rendering with Word/Glyph/WordOrGlyph wrap modes |
+| `undo_redo`        | Undo and redo traversal at varying history depths |
+
+## Filtering within a target
+
+Pass a filter string after `--` to run only matching cases:
 
 ```sh
-cargo bench append::1_lorem
+cargo bench -p tui-textarea-bench --bench insert_append -- append::10_lorem
 ```
 
-To compare benchmark results with [critcmp][]:
+## Comparing branches with [critcmp][]
 
 ```sh
 git checkout main
-cargo bench -- --save-baseline base
+cargo bench -p tui-textarea-bench -- --save-baseline base
 
 git checkout your-feature
-cargo bench -- --save-baseline change
+cargo bench -p tui-textarea-bench -- --save-baseline change
 
 critcmp base change
 ```
 
+## Note on `cargo test -p tui-textarea-bench`
+
+Running `cargo test` on this package is a smoke test only — it verifies
+that the benchmark code compiles and runs without panicking, but produces
+no timing data. Use `cargo bench` or the Rust runner for real measurements.
+
 [criterion]: https://github.com/bheisler/criterion.rs
 [critcmp]: https://github.com/BurntSushi/critcmp