Don't propagate errors into Node.js

[armanbilge]

H/t @satabin for reporting in gnieh/fs2-data#335 (comment).

Previously, exceptions encountered in fs2.Stream would also be propagated into the Node.js stream it was interopping with. In v3.2.8 this started crashing the Node.js process with:

node:events:505
      throw er; // Unhandled 'error' event
      ^
Error: unexpected '1' before object key
    at $c_Lfs2_io_internal_ThrowableOps$ThrowableOps.toJSError__sjs_js_Error (/home/runner/work/fs2-data/fs2-data/json/circe/.js/target/scala-2.13/fs2-data-json-circe-test-fastopt/main.js:11048:10)
    at /home/runner/work/fs2-data/fs2-data/json/circe/.js/target/scala-2.13/fs2-data-json-circe-test-fastopt/main.js:11134:92
    ...

(Note: the unexpected '1' before object key is an expected error raised in the fs2-data test suite.)

Propagating the error seems sensible, but probably is not the right thing to do.

1. It's not necessary to preserve the error, since it's already being handled primarily through the FS2/CE3 control flow, propagation would simply be a side-channel.
2. The error is likely to be unhandled on the Node.js side, which means it will crash the process. This effectively circumvents FS2/CE3 control flow which is undesirable.

I think this was an existing bug, that became exposed through the changes in #2918. Specifically in 0c71749 besides replacing the facade I swapped in the MicrotaskExecutor for SyncIO (probably should have done it in another PR :) This introduced a new cede/yield due to the evalOn which gave the error event an opportunity to emit when previously it did not.