Simplify the generated files, move testthat functionnality in external script

[vogr]

After this PR, the generated files are much simpler. Here is for instance yaml/as.yaml/test-36.R:

genthat_extracted_call <- function() {
    yaml:::as.yaml(x = 0)
}

The return value and the seed are recorded in the .ext file (with the function record_test_exts) after the generation of the file ; this way the files can be run in isolation (using callr).

This makes it possible to still use the generated files as unit tests using testthat, but also to use them for other uses like benchmarks. These functionalities are taken care of by external scripts. This approach has the advantage that it easier to add new functionalities without having to re-generate all the files.

For instance, the files benchmark.R and test_file.R (in tools/harnesses/) implement the previously mentioned functionalities:

# with the tests generated in ./tests

# Test one file
$ "/path/to/harnesses/test_files.R" "tests/yaml/as.yaml/test-36.R"
Test passed 🎉

# Test all the files in "tests"
$ GENERATED_TESTS_DIR="./tests"  "/path/to/harnesses/test_files.R"
Test passed 🎉
Test passed 😀
Test passed 🥳

# From R with test_file (pass environment variable to R, or use withr::with_envvar)
$ GENERATED_TESTS_DIR="/path/to/tests" R
> testthat::test_file("/path/to/harnesses/test_files.R")
══ Testing test_files.R ═════════════════════════════════════════════════════════
[ FAIL 0 | WARN 0 | SKIP 0 | PASS 3 ] Done!

# Benchmark the test (the output format is compatible with [ReBench](https://rebench.readthedocs.io))
$ benchmark.R tests/yaml/as.yaml/test-36 5 100
tests/yaml/as.yaml/test-36: iterations=1 runtime: 3876us
tests/yaml/as.yaml/test-36: iterations=1 runtime: 1234us
tests/yaml/as.yaml/test-36: iterations=1 runtime: 1222us
tests/yaml/as.yaml/test-36: iterations=1 runtime: 1226us
tests/yaml/as.yaml/test-36: iterations=1 runtime: 1189us
tests/yaml/as.yaml/test-36: iterations=5; average: 1749 us; total: 8747us