In this PR we decouple the regjsparser switch from the regexpu-core transformer switch. It is a breaking change since it does not throw for \\p{Unified_Ideograph} when unicodePropertyEscape is not enabled, instead it assumes our user know exactly what she is doing and only transform the features as instructed.

By doing this, babel-preset-env can have more granular control on ES6+ regex features. For example, when user specifies

{
  "target": "Hypothetical browser supports unicode property escape but not named capturing groups"
}

and let's say the pattern we want to transform is /(?<name>\p{Unified_Ideograph}{3})/u we don't have to transform unicode property escape here and instead should only pass namedGroup: true.

rewritePattern(
  /(?<name>\p{Unified_Ideograph}{3})/u,
  "u",
  { namedGroup: true }
)
// expect to transform `?<name>` only without touching `\p{Unified_Ideograph}`

However, to transform this pattern, we still need regjsparser support for unicode property escape otherwise the parser will throw on \p{. Therefore, we maximize the parser support, allow parser to parse any new syntax, and only transform when user opt-in the transformation. By doing so we encourage to ship native browser features without superfluous transformations.

Edits: the previous Safari 12 example above is practically invalid as of Sep 2019. Every browser supports unicode property escape also support named capturing groups.