Metric: Robustness lints

[dgkf]

Today we discussed adding lintr's robustness lints, which are lints for relatively unopinionated good programming practices that are potentially error-prone.

As far as implementation goes, we considered two approaches:

1. Either calculate all lints internally, then surface robustness lints as a metric. This allows extension packages to re-use this data for the purpose of surfacing more opinionated lints.
2. Calculate only robustness lints, which should be significantly faster, but would require that for more opinionated use cases, all lints would have to be re-executed.

We decided to go with (2) because running all lints can be quite costly for large code bases and we wouldn't be able to handle the case where organizations might even have custom lints that they wish to add, so (1) might require re-execution anyways.

[llrs-roche]

Note that some of these are in this category that mentions some concerns about security: https://lintr.r-lib.org/reference/executing_linters.html. Just pointing this so that we consider our defaults for when people run this on internal servers.

[dominik-appsilon]

I started a draft PR which contains working implementation of some version of this metric. Again, I encountered multiple decision points and raised all in the discussion in the aforementioned PR.