In `@crates/vm/src/stdlib/warnings.rs`:
- Around line 11-17: The call to crate::warn::warn passes stack_level as
(stack_level as isize) which can wrap for very large usize values; change this
by performing a safe conversion from usize to isize (e.g., use isize::try_from
or usize::try_into) and handle the failure path explicitly (either clamp to
isize::MAX or return an error/early-return) before calling crate::warn::warn;
update the call site that constructs the args (the stack_level conversion in the
warn invocation and any surrounding logic that sets stack_level) so it uses the
checked conversion result and passes the resulting isize safely to
crate::warn::warn.

In `@crates/vm/src/warn.rs`:
- Around line 249-256: In normalize_module, the code uses filename.char_len() to
compute a slice index into filename.as_wtf8(), which mixes character count with
byte slicing and breaks for non-ASCII names; change to use filename.byte_len()
(or similar byte-length API) and use that byte length when computing the slice
(e.g., replace the char_len() binding with a byte_len variable and slice
as_wtf8()[..byte_len - 3]) so the ".py" strip uses byte offsets consistent with
as_wtf8(); keep the existing ends_with(b".py") check and only adjust the length
source in normalize_module.

In `@crates/vm/src/warn.rs`:
- Around line 288-292: The code currently silences conversion errors by using
tmp_item[4].try_int(vm).map_or(0, ...) to set ln, which masks malformed filter
configs; change this to propagate the integer conversion error instead of
defaulting to 0 (e.g., use tmp_item[4].try_int(vm)? and then convert to
usize/isize as appropriate, or map_err to return the VM error), so that failures
in tmp_item[4].try_int(vm) bubble up rather than making ln == 0; update any
downstream handling of ln (used in the if condition with is_subclass,
check_matched, lineno) to expect a fallible conversion.