Pull Request Description

Makes safepoints use a volatile boolean flag instead of an assumption. This results in 0%-40% slowdowns across our benchmark suites (where the impact becomes smaller, the more memory is involved). This should be revised when a better safepoint mechanism makes it into Truffle.

Important Notes

The safepoints are moved around – the approach we've taken so far ("safepoint on every method call") is not
sustainable with this solution (resulting in benchmarks becoming even 3x slower). Therefore safepoints are inserted:

1. At the start of non-builtin non-inlined Enso methods.
2. At the beginning of each @TailCall loop iteration.
   This may turn out to be too sparse, in which more poll calls can be sprinkled.

Checklist

Please include the following checklist in your PR:

• The documentation has been updated if necessary.
• All code conforms to the Scala, Java, and Rust style guides.
• All documentation and configuration conforms to the markdown and YAML style guides.
• All code has been tested where possible.