windows: fix vectored exception handler lifecycle for repeated thread env init/destroy

[kogum4]

Summary

This PR fixes Windows VEH lifecycle handling for repeated runtime/thread-env init-destroy cycles.

On Windows host integrations that repeatedly unload/reload WASM runtime users, stale VEH state could remain and later crash in ntdll exception dispatch.

Reproduction (before)

On Windows, in a host-like lifecycle that repeatedly creates/destroys runtime usage contexts (including thread env init/destroy churn), intermittent crashes were observed in exception dispatch (ntdll, access violation path) after unload/reload cycles.

Changes

File:

• core/iwasm/common/wasm_runtime_common.c

What changed:

• add Windows-only VEH ref count
• guard VEH lifecycle updates with a mutex
• register VEH only when ref count transitions 0 -> 1
• remove VEH only when ref count transitions 1 -> 0
• if RemoveVectoredExceptionHandler fails, keep retry-able ownership state instead of dropping handler state

Behavior change

• Before: VEH lifecycle could become inconsistent under repeated init/destroy cycles, leading to intermittent post-unload crash behavior.
• After: VEH registration/removal is reference-counted and synchronized, with robust failure handling on removal.

Validation

• rebuilt WAMR on Windows
• rebuilt downstream integration
• passed downstream C++ tests
• reran host-style repeated unload/reload + open/close loop; crash no longer reproduced in the previously failing scenario

Notes

No new WAMR-internal test is included in this patch; validation was performed in the integration lifecycle where the issue reproduces.
If maintainers prefer, I can submit a follow-up PR with a WAMR-side regression test for this lifecycle pattern.

[TianlongLiang]

This PR overall LGTM with one minor comments