fix: POSIX :fcntl_h seek constants + cleaner Java-exception error reporting

[fglock]

Summary

Two small, independent runtime/stdlib fixes surfaced while investigating jcpan -t IO::Any.

POSIX :fcntl_h now exports SEEK_SET / SEEK_CUR / SEEK_END

Real perl's POSIX :fcntl_h exports these; ours did not. Modules that do
use POSIX qw(:fcntl_h) and then use SEEK_SET as a bareword — notably
File::Slurp — die at runtime with Bareword "SEEK_SET" not allowed while "strict subs" in use at File/Slurp.pm line 119, which cascades into
IO::Any, Config::Simple, and anything else pulling in File::Slurp.

CLI reporter no longer surfaces raw Java exception class names

When a Java exception leaked through to the reporter, users would sometimes
see two cryptic Java-looking lines above the real message:

java.lang.NoClassDefFoundError: org/perlonjava/runtime/operators/KillOperator
org.perlonjava.runtime.operators.KillOperator
        TAP::Parser::Iterator::Process at jar:PERL5LIB/TAP/Parser/Iterator/Process.pm line 272

Two causes:

• A wrapping exception whose message was built from inner.toString()
  ("java.lang.Foo: msg") caused both the wrapper and the inner message to
  print.
• NoClassDefFoundError (slash-separated) commonly wraps
  ClassNotFoundException (dot-separated) for the same class; the two
  differ only in / vs . and were treated as independent messages.

ErrorMessageUtil.stringifyException gains two helpers:

• isJavaToStringOf(outer, inner) — suppresses the outer line when it is
  literally Throwable.toString() of any ancestor in the cause chain, or
  differs only in / vs . from an ancestor's message.
• stripJavaExceptionPrefix(message) — strips a leading fully-qualified
  Java exception class prefix (java.lang.X: …, org.foo.YException: …).
  The pattern requires a dotted prefix with at least two segments plus
  a recognizable Exception / Error / Throwable suffix, so legit Perl
  messages like "DBI error: handle closed" or "Foo: not a java class"
  pass through unchanged.

Impact on jcpan -t IO::Any

Test file	Before	After
t/01_IO-Any.t	15/34 failed, died at test 20 (SEEK_SET)	1/34 failed (platform flock)
t/02_IO-Any_AnyEvent.t	timed out, noisy NoClassDefFoundError output	timed out, clean output
t/03_DATA.t	ok	ok

Remaining failures are unrelated to this PR:

• t/01_IO-Any.t test 30 is a non-blocking shared-lock assertion that hits
  EDEADLK on macOS — flock emulation quirk.
• t/02_IO-Any_AnyEvent.t times out because AnyEvent::Loop's I/O watchers
  are destroyed on return (a push @q, $self; weaken $q[-1] refcount issue
  that is being addressed on a separate refcount branch).

Test plan

• Added ErrorMessageUtilTest with 5 cases (wrapped NCDF, slashed/dotted
  dedup, Perl pass-through, false-positive Foo: bar strings).
• make — full unit test suite green.
• Manual re-run of jcpan -t IO::Any before/after; output diff above.

Generated with Devin