Adopt wil/resource.h (wil::unique_*) for Win32 handles across the codebase

[asklar]

Goal

Adopt wil/resource.h RAII wrappers for raw Win32 handles across the codebase, replacing manual CloseHandle/FreeLibrary/DeleteObject/RegCloseKey/etc. This complements the COM-pointer cleanup (#10) — same "use WIL RAII" theme, handles instead of interfaces.

Do it as one PR once #23 (WinForms) and #24 (COM ptrs) have landed (so it also covers src/tap_winforms/winforms_tap_native.cpp and doesn't collide with #24's TAP edits).

Scope (audit: ~160 sites across ~14 files)

Map raw handles to the appropriate wil::unique_*:

• Generic HANDLE (named pipes, events, files, CreateToolhelp32Snapshot) → wil::unique_handle / wil::unique_hfile / wil::unique_event
• Process/thread handles (OpenProcess, CreateProcess PROCESS_INFORMATION) → wil::unique_process_handle / wil::unique_handle
• HMODULE from LoadLibrary → wil::unique_hmodule
• Registry HKEY → wil::unique_hkey
• GDI objects in screenshot.cpp (HDC/HBITMAP/HFONT/HPEN/HBRUSH) → wil::unique_hdc / wil::unique_hbitmap / wil::unique_hgdiobj

Files (raw-handle site counts)

• src/plugin_avalonia/lvt_avalonia_plugin.cpp (~22), avalonia_tap_native.cpp (~7)
• src/providers/wpf_inject.cpp (~21), xaml_diag_common.cpp (~21), winui3_provider.cpp (~4), comctl_provider.cpp (~2)
• src/plugin_chromium/chromium_host.cpp (~20), lvt_chromium_plugin.cpp (~19)
• src/tap_wpf/wpf_tap_native.cpp (~12), src/tap/lvt_tap.cpp (~5), src/tap_winforms/winforms_tap_native.cpp (after Add WinForms provider (+ sample app) #23)
• src/plugin_loader.cpp (~4), src/target.cpp (~2), src/framework_detector.cpp (~1)
• src/screenshot.cpp (GDI objects)

Caveats

• TAP DLLs (lvt_tap, wpf_tap_native, winforms_tap_native, avalonia_tap_native) use static CRT (/MT) and avoid C++ exceptions — use the non-throwing WIL forms there; do not introduce throwing helpers.
• Watch for handles whose ownership is intentionally transferred (e.g. a named-pipe handle handed to a fire-and-forget worker thread, or handles closed on another thread) — don't wrap in a way that double-closes or closes early.
• Behavior must be identical; this is a leak-hygiene/RAII refactor, not a functional change. Keep build\lvt_unit_tests.exe + integration tests green.

Relates to #10.

[asklar]

Audit note: non-COM pointer/allocation review (for scoping)

Audited raw allocations across src/ while scoping this issue:

Add to this issue''s scope (BSTR / CoTaskMem in the TAP DLL):

• src/tap/lvt_tap.cpp: SysFreeString(...) on PropertyChainValue/SourceInfo BSTR fields, and CoTaskMemFree(props/sources) for the arrays returned by GetPropertyValuesChain. Map to wil::unique_bstr / wil::unique_cotaskmem where feasible.
  • Caveat: these are fields inside API-returned struct arrays and live in the SEH/thread-affinity path — RAII with destructors can''t be used inside __try frames, so prefer wil::scope_exit for the array cleanup or leave manual with a comment where SEH-constrained.

Explicitly OUT of scope (leave as-is):

• src/plugin_avalonia/lvt_avalonia_plugin.cpp and src/plugin_chromium/lvt_chromium_plugin.cpp: malloc/free of the result string is the plugin C ABI — the plugin allocates, the host frees via a C function pointer across the DLL boundary. Must NOT become std::unique_ptr/wil (different CRT heaps).

No std::unique_ptr/std::shared_ptr migration needed: the remaining new/delete are COM-refcount patterns (delete this, new LvtTap()), and the object model already uses value semantics (std::vector<Element>, std::string) + RAII (RemoteBuffer).

[asklar]

Correction re: SEH caveat — BSTR/CoTaskMem cleanup is straightforward

Verified the enclosing functions: the SysFreeString/CoTaskMemFree sites are in pure-C++ functions (CollectBoundsForNode, SetSiteImpl), while the __try/__except is isolated in thin wrappers (CollectBoundsForNodeSEH, SetSite) — the code already follows the "no C++ objects with destructors inside __try" rule.

So there is no SEH blocker at those sites:

• Wrap each PropertyChainValue/SourceInfo BSTR field in wil::unique_bstr (adopt the field, auto-free at scope end).
• Wrap the props/sources arrays from GetPropertyValuesChain in wil::unique_cotaskmem (freed after the per-field BSTRs).

wil::scope_exit is not required here. (Still use non-throwing WIL forms given the TAP DLL's /MT, no-C++-exceptions policy.)

[asklar]

⤵️ Moved to #38 — this issue stays scoped to resource RAII

The try_query / WIL result-macro (RETURN_IF_FAILED) modernization has been split out into #38 to keep this PR focused on wil::unique_* handles + wil::unique_bstr/wil::unique_cotaskmem (resource RAII).

Rationale: handles/BSTRs are real leak-hygiene fixes; the try_query/result-macro pass is pure readability polish with no behavior change. Separating them keeps each PR cohesive and independently reviewable/validatable.

This issue (#37) = handles + BSTR/CoTaskMem only. See #38 for the COM/HRESULT idiom polish.