Address CRAN package check macOS issues

[riccardoporreca]

Before submitting a new version of rTRNG to CRAN, we should address the macOS errors on the CRAN package check page.

In particular, the following issues are reported

Some tests are failing on macOS

══ testthat results ═══════════════════════════════════════════════════════════
[ OK: 581 | SKIPPED: 0 | WARNINGS: 0 | FAILED: 30 ]
1. Failure: $split errors for out-of-range subsequence indices (@test-TRNG.Engine.R#264)
...

In fact we have exactly 30 warnings when running tests in Ubuntu for the same test reported here

test-TRNG.Engine.R:264: warning: $split errors for out-of-range subsequence indices
`e$split(p, 0L)` generated an S3 error and you are testing the error message.
* The error has class = c("std::invalid_argument", "C++Error", "error", "condition")
* Testing with `class` is more robust than testing with `regexp`.
* Do you want `expect_error(..., class = "std::invalid_argument")`?

Most likely this is detected as a failure under macOS, and fixing the expectation as hinted in the warning message might solve the issue.

Re-building the vignettes fails on macOS (and Solaris)

This happens for C++ code chunks using sourceCpp()

Library not loaded: rTRNG.so

[riccardoporreca]

• Making sure no warnings are triggered by the unit tests committed with e4ac407

• MacOS checks enabled on TravisCI with e529adf, revealing the issues with building vignettes

• Issues with building / linking against rTRNG in C++ code chunks (and similarly in packages) addressed by conditionally evaluating them based on a check for a minimal sourceCpp() example => d9e24dd fixes the issues macOS issues on Travis. The general approach is to keep this optional and document it properly.

  • => documentation done in ab1c511481641640e7d9f057ea50d90312d1b596