fix: handle StorageFull and I/O errors in Java world save instead of panicking - #787

…panicking

Replace all .expect()/.unwrap() calls in create_region, save_single_region,
and save_java with proper error propagation. A Mutex captures the first error
from the rayon parallel iterator; save() detects disk-full conditions and
calls emit_gui_error() with a clear user-facing message instead of crashing.

… create_base_chunk Result, correct disk-full detection

- Replace lock().unwrap() with lock().unwrap_or_else(|p| p.into_inner()) on
  all three Mutex accesses so a poisoned mutex never causes a secondary panic
- Add AtomicBool should_stop for lock-free fast-path early exit in the rayon
  loop; the Mutex is now only acquired when actually storing the error value
- Change create_base_chunk to return Result and propagate the fastnbt
  serialization error with ? instead of unwrap, closing the last panic hole
- Replace inline string-based disk-full detection with is_disk_full_error()
  helper that walks the full error chain via .source(); fixes the 'code: 112'
  pattern which never matched Windows io::Error Display ('os error 112')

…isk_full_error correctness

Comment 2 (is_disk_full_error docstring/implementation mismatch):
- Walk source chain via .source() which yields &(dyn Error + 'static),
  enabling downcast_ref::<io::Error>() for structured checks
- Check io_err.kind() == ErrorKind::StorageFull and raw_os_error() 112/28
  as primary detection; keep Display string matching as fallback for
  wrappers (e.g. fastanvil) that don't expose io::Error in source chain
- Fix docstring to accurately describe the hybrid approach

Comment 3 (HIGH - genuine regression: save() swallowed errors silently):
- Change save() from void to Result<(), Box<dyn Error + Send + Sync>>
- Return Err(e) after emitting the GUI error, so callers see the failure
- In data_processing.rs, propagate with if let Err(e) = editor.save() { return Err(...) }
- Previously: disk-full showed GUI error but pipeline continued to
  'Finalizing world...' and returned Ok(output_path) - frontend saw success
  Now: pipeline aborts cleanly on save failure

Skipped comment 1 (path context nit - no anyhow dependency, boilerplate cost
outweighs benefit for user-facing errors that already include OS message)
Skipped comment 4 (try_for_each style nit - AtomicBool approach is correct
and has a marginally better fast-path than try_for_each)