optim: add assumptions to Babel config, mirroring previous config

[agilgur5]

Summary

Optimize output by adding assumptions to the Babel config, which mirrors the previous build system's config that uses Babel 7.

• By previous, I mean the several year old one prior to TypeScript Rewrite #42

Details

• previously used Babel 6's preset-stage-2, which, in Babel 7, is the equivalent of turning on loose in proposal-class-properties:
  https://github.com/babel/babel/tree/v7.5.5/packages/babel-preset-stage-2

• top-level assumptions replace the old loose configurations:
  https://github.com/babel/rfcs/blob/main/rfcs/0003-top-level-assumptions.md#assumptions-list

• this makes the minified UMD size ~20% smaller, very close to the size it was previously w/ Babel 6

  • basically the size increase was, in part, due to added spec-compliancy cases
    • and we don't actually need that stricter spec-compliancy in this codebase's small usage, so we can add "assumptions" to have the compiler optimize more by skipping some checks / edge cases that are unnecessary in this case

• as a double-check measure, I compared the old package's minified UMD to see if it had the helpers in it

  • this was a bit hard to do as it was minified after all, but I was able to more-or-less correlate all the helpers with either error messages in the helper code (which can't be minified), or specific methods or properties that aren't normally touched (e.g. enumerable)

    • prev minified UMD: https://unpkg.com/react-signature-canvas@1.0.5/build/index.js
    • Babel helper code: https://github.com/babel/babel/blob/a24c1d047e6db77c32d9505c6876897b4c3689d2/packages/babel-helpers/src/helpers.ts#L207

  • this looks be a good bit closer to the old build's results, but should still be wary of possible legacy compat issues and note that in the release notes

    • in particular, the implicit update to compat-table that happened during upgrading of the build system, which would change what defaults means for browserslist etc

History

I knew the output was different (and larger) due to spec-compliancy changes, but somehow didn't think about adding loose etc as an optimization until I saw Babel's assumptions RFC. I thought I wrote about the output differences in #42 (review) , but I guess I didn't? Couldn't find it in a commit message either, but could've sworn I wrote it down somewhere...

[agilgur5]

LGTM. Tiny change. Tests pass, and I double-checked that the example still works when using the UMD build too