fix(encode,errno): add iso-10646-1 alias and make local $! reset errno

[fglock]

Summary

Investigation of jcpan -t CSV::Processor revealed two related PerlOnJava bugs that cascaded through the dependency chain (CSV::Processor → Text::AutoCSV → File::BOM). Both are fixed here in a single commit.

1. Encode.java: missing iso-10646-1 alias

Real Perl's Encode module defines iso-10646-1 as an alias for UCS-2. PerlOnJava had aliases for UCS-2, UCS-2BE, UCS-2LE but not iso-10646-1 / ISO-10646-1.

File::BOM references this alias when populating its enc2bom table, so use File::BOM died at compile time with Unknown encoding: iso-10646-1. The cascading effect was that everything depending on File::BOM (Text::AutoCSV, etc.) also failed to load.

2. ErrnoVariable.dynamicSaveState: local $! was a no-op

local $!; is supposed to clear $! to the empty/no-error state inside the dynamic scope (and restore the prior value on scope exit). The base RuntimeScalar.dynamicSaveState resets type/value to UNDEF, but ErrnoVariable overrides the readers (toString, getInt, getBoolean, ...) to return its private errno/message fields, which were not being cleared during local. Result: local $!; saved the old value but did not hide it, so any subsequent read still returned the stale errno.

This breaks the very common idiom used by File::BOM's _safe_read:

local $!;
my $status = read($fh, my $out, $count);
die $! if !$status && $!;

A stale $! from earlier in the program made every legitimate end-of-file read look like an error.

Minimal repro (before fix):

$ ./jperl -e '$! = 1; sub f { local $!; print "inside: $!\n" } f()'
inside: Operation not permitted    # bug — should be empty

After fix it prints inside: (empty), matching system Perl.

Test impact (running with ./jperl)

Module	Before	After
File::BOM t/01..bom.t	dies after 41 tests	85 tests pass before unrelated POSIX::mkfifo cutoff
Text::AutoCSV (entire suite)	7/48 (cascade load failure)	535/560 (95.5%)
CSV::Processor	blocked at load time	loads; Utils.t passes

Test plan

• make (full unit-test suite) passes locally
• local $! smoke test matches system Perl
• File::BOM t/01..bom.t reaches 85 passing tests (was 41)
• Text::AutoCSV regains 535/560 passing tests
• No other behavior changes — dynamicRestoreState already restores the saved errno/message from the existing stacks

Remaining issues (out of scope, separate tickets)

• LWPx::TimedHTTP test suite uses fork (unimplemented in PerlOnJava); the eval { fork } SKIP guard doesn't trigger correctly and tests hang/fail. Blocks Email::Extractor, which in turn blocks CSV::Processor runtime tests.
• File::BOM t/02..perlio-via.t uses PerlIO::via, already documented as unimplemented in dev/modules/perlio_via.md.
• File::BOM 00..setup.t uses POSIX::mkfifo, also unimplemented (would let the remaining 30 tests run if available).
• Email::Extractor has an undeclared Mail::Address dependency in its META — that's a CPAN module bug, not PerlOnJava.
• CSV::Processor::Utils.pm emits a "my" variable $path masks earlier declaration warning — CPAN module's own code style.

Generated with Devin