Stop find_c aborting on directory entries

[pathanin]

Fixes a macOS-only crash: opening the build list with a sub-folder under Builds/ throws

In 'OnFrame': C++ exception:
filesystem error: in file_size: Is a directory [".../src/Builds/testpob"]
stack traceback:
[C]: in function 'NewFileSearch'
Modules/BuildListHelpers.lua:63: in function 'ScanFolder'

Cause

find_c::FindFirst/FindNext called iter->file_size() — the throwing overload — on every entry that matched the glob. directory_entry::file_size() has no defined result for a directory: MSVC returns the size cached by FindFirstFile (0) and never throws, libc++ calls stat and throws. So the code is correct on Windows and fatal everywhere else.

The throw happens in find_c, before NewFileSearch's findDirectories filter at ui_api.cpp:1354, so it is not limited to directory searches — any NewFileSearch(dir.."*") dies if dir holds a sub-directory. Affected call sites in PoB-PoE2: Modules/BuildListHelpers.lua:63, Classes/FolderListControl.lua:46,82, Classes/BuildListControl.lua:144, Modules/Main.lua:1583-1634 (folder copy/delete walks), Export/Main.lua:712-763.

Change

Query metadata through the error_code overloads and report 0 for a directory — the value Windows already produces, so no caller sees a behaviour change. is_directory() and last_write_time() get the same treatment: they fail out of the same status() call, so an unreadable or dangling entry aborted the process by the identical route.

Verification

Windows CI cannot cover this — the divergence is libc++ vs MSVC. Verified locally against the shipped arm64 dylib with a probe that calls find_c::FindFirst/FindNext directly over src/Builds/, which contains one sub-folder (testpob) and two files:

dylib	result
before	RESULT: THREW - filesystem error: in file_size: Is a directory [.../Builds/testpob] on the first entry
after	testpob dir=1 size=0, .DS_Store dir=0 size=8196, ~~temp~~.xml dir=0 size=53075; *.xml glob returns 1; recursing into testpob/* returns 2; no exception

Built with the standard arm64-osx-dynamic config; incremental relink, LuaJIT untouched so the soname is unchanged. runtime/libSimpleGraphic.dylib restaged with @loader_path re-added; all 9 @rpath deps resolve. Backup at ~/Documents/pob-runtime-backup-2026-07-25/.

A failed last_write_time() returns file_time_type::min(), whose negative count would wrap to a huge value in find_c::modified (unsigned long long) and sort the entry to the top of the build list's "Sort by Last Edited"; that reports 0 instead.

[pathanin]

End-to-end confirmed on the staged arm64 runtime: the build list renders the testpob folder row, opens it, and lists both builds inside. That closes the one gap the find_c probe could not cover.