Fix #415: Report usage positions of missing definitions when linking

[WojciechMazur]

This PR adds aggregation of missing definitions with their usage positions when performing static reachability phase.
Currently, when some definition is missing, we would result in a single exception due to usage of Map.apply which does not contain any greater context.
Thanks to this PR we would get list of all missing definitions with places of their usage, eg.:

    [error] Found 2 missing definitions while linking
    [error] Not found Top(java.time.Instant$)
    [error] 	at <project_dir>/javalib/src/main/scala/java/nio/file/attribute/FileTime.scala:9
    [error] Not found Top(java.util.Locale$)
    [error] 	at <project_dir>/javalib/src/main/scala/java/lang/String.scala:11
    [error] 	at <project_dir>/javalib/src/main/scala/java/lang/String.scala:43
    [error] 	at <project_dir>/javalib/src/main/scala/java/text/DateFormatSymbols.scala:28
    [error] 	at <project_dir>/javalib/src/main/scala/java/text/DecimalFormat.scala:60
    [error] 	at <project_dir>/javalib/src/main/scala/java/util/Formatter.scala:6
    [error] 	at <project_dir>/javalib/src/main/scala/java/util/Formatter.scala:45
    [error] 	at <project_dir>/javalib/src/main/scala/java/util/Formatter.scala:53
    [error] stack trace is suppressed; run last tests / Test / nativeLink for the full output
    [error] (tests / Test / nativeLink) scala.scalanative.linker.Reach$ReachabilityFailureException: Undefined definitions found in reachability phase

Resolves #415

[sjrd]

This is cool :)

[sjrd]

I don't see a reason for NoStackTrace here. NoStackTrace is useful for exceptions thrown in non-failure paths, which must therefore be fast to initialize. This is not the case here.

Perhaps you wanted to avoid the stack trace to be displayed in the sbt shell? If that's the case, then there is a better solution: the sbt plugin should catch the exception and wrap it in a MessageOnlyException.

[sjrd]

Also, consider extending Exception rather than RuntimeException. Or better: don't we have a generic LinkingException somewhere? If not, maybe we should define one.