Split test-examples into two phases: build all examples in parallel
first, then run tests serially against pre-built binaries. Abort on any
build failure instead of skipping tests for that example.

Replace the centralized `examples/test-examples.toml` with individual
`test.toml` files in each example directory, updating the test runner
and related documentation accordingly.

Replace the separate code/docs CI jobs and local check flags with
a unified `--check-*` flag system, enabling each check to run
independently across a platform matrix.

Apply project-specific color coding (0-50% red, 51-80% yellow, 81-100%
green) to the per-file coverage summary in the generated HTML report,
overriding llvm-cov's default thresholds.

Replace `cargo llvm-cov run` with explicit RUSTFLAGS-compiled examples
to ensure dependency crates are instrumented, fixing silent coverage
loss (e.g., once_exec.rs at 0%).

Refactor cov-test to collect coverage from workspace tests, integration
test crates, and examples, merging them into a single HTML report.

This requires the forked cargo-llvm-cov which supports including
non-workspace binaries and test/example source files in the report.

Prefer both streams when present so failures with stderr
warnings still include stdout details like test assertions.

Avoid ProgressBar swallowing error output in CI or piped contexts by
printing directly to stdout when stdout is not a terminal.

Replace the ZIP-based template archive workflow with direct
directory copying. Delete the `package-mling-tmpls` tool and
remove the `zip` dependency.