Improve diagnostics when the packager cannot be reached or used

[javache]

Summary:
Three gaps made an unusable packager connection close to impossible to
diagnose from logs.

ReconnectingWebSocket logged only Couldn't connect to "<url>", will silently retry, discarding the Throwable that onFailure already
receives. That throwable is the only thing separating a server which
rejected the WebSocket upgrade from one that was never reached — okhttp
reports the former as Expected HTTP 101 response but was '<code> <message>'
— so without it the two produced byte-identical output. It is now logged, and
onClosed distinguishes a peer-side close by including its code and reason.

ReactHostImpl silently falls back to the bundle shipped inside the app when
useDevSupport is true but allowPackagerServerAccess is false. No packager
request is issued in that state, so nothing downstream has anything to
report, and the app appears to ignore a perfectly healthy development server
for no visible reason. That combination now warns; the existing debug log is
kept for the ordinary case where dev support is off too.

Also repairs a comment in ReactHostImpl that had unrelated text spliced
into the middle of a word.

No public API changes, so ReactAndroid.api is unaffected.

Changelog:
[Android][Fixed] - Log the underlying cause when a packager connection fails
[Android][Added] - Warn when dev support is enabled but packager server access is disabled

Differential Revision: D119721144

[meta-codesync]

@javache has exported this pull request. If you are a Meta employee, you can view the originating Diff in D119721144.