Fixes and tests

line_profiler/_line_profiler.pyx
    _ThreadState
        - Renamed from `ThreadState`
        - Made `._wrap_trace` unaccessible on the Python level
    LineProfiler
        wrap_trace.__set__()
            Now syncing between `_ThreadState`s in `._all_thread_states`
        _thread_state.__get__()
            If another thread is already initialized, get the
            `wrap_trace` therefrom

tests/test_sys_trace.py
    New test module (mostly cherry-picked from before we merged pyutils#347):
    - `test_callback_preservation()`
    - `test_callback_wrapping()`
    - `test_wrapping_throwing_callback()`
    - `test_wrapping_line_event_disabling_callback()`
    - `test_wrapping_thread_local_callbacks()`

Author: TTsangSC

line_profiler/_line_profiler.pyx

@@ -301,7 +301,7 @@ list[tuple[int, int, int]]]):
         self.unit = unit
 
 
-cdef class ThreadState:
+cdef class _ThreadState:
     """
     Helper object for holding the thread-local state; documentations are
     for reference only, and all APIs are to be considered private and
@@ -310,7 +310,7 @@ cdef class ThreadState:
     cdef TraceCallback *legacy_callback
     cdef dict mon_callbacks  # type: dict[int, Callable | None]
     cdef public object active_instances  # type: set[LineProfiler]
-    cdef public int _wrap_trace
+    cdef int _wrap_trace
 
     def __init__(self, instances=(), wrap_trace=False):
         self.active_instances = set(instances)
@@ -534,7 +534,7 @@ cdef class LineProfiler:
     cdef public object threaddata
 
     # These are shared between instances and threads
-    # type: dict[int, set[LineProfiler]], int = thread id
+    # type: dict[int, _ThreadState], int = thread id
     _all_thread_states = {}
     # type: dict[bytes, int], bytes = bytecode
     _all_paddings = {}
@@ -694,12 +694,17 @@ cdef class LineProfiler:
             self.threaddata.enable_count = value
 
     # These two are shared between instances, but thread-local
+    # (Ideally speaking they could've been class attributes...)
 
     property wrap_trace:
         def __get__(self):
             return self._thread_state.wrap_trace
         def __set__(self, wrap_trace):
-            self._thread_state.wrap_trace = wrap_trace
+            # Make sure we have a thread state
+            state = self._thread_state
+            # Sync values between all thread states
+            for state in self._all_thread_states.values():
+                state.wrap_trace = wrap_trace
 
     property _thread_state:
         def __get__(self):
@@ -708,16 +713,23 @@ cdef class LineProfiler:
                 return self._all_thread_states[thread_id]
             except KeyError:
                 pass
-
-            # First instance, load default `wrap_trace` value from the
-            # environment
-            # (TODO: migrate to `line_profiler.cli_utils.boolean()`
-            # after merging #335)
-            from os import environ
-
-            env = environ.get('LINE_PROFILE_WRAP_TRACE', '').lower()
-            wrap_trace = env not in {'', '0', 'off', 'false', 'no'}
-            self._all_thread_states[thread_id] = state = ThreadState(
+            # First profiler instance on the thread, get the correct
+            # `wrap_trace` value and set up a `_ThreadState`
+            try:
+                state, *_ = self._all_thread_states.values()
+            except ValueError:
+                # First thread in the interpretor: load default
+                # `wrap_trace` value from the environment
+                # (TODO: migrate to `line_profiler.cli_utils.boolean()`
+                # after merging #335)
+                from os import environ
+
+                env = environ.get('LINE_PROFILE_WRAP_TRACE', '').lower()
+                wrap_trace = env not in {'', '0', 'off', 'false', 'no'}
+            else:
+                # Fetch the `.wrap_trace` value from an existing state
+                wrap_trace = state.wrap_trace
+            self._all_thread_states[thread_id] = state = _ThreadState(
                 wrap_trace=wrap_trace)
             return state
 
@@ -909,7 +921,7 @@ cdef extern int legacy_trace_callback(
        https://github.com/python/cpython/blob/de2a4036/Include/cpython/\
 pystate.h#L16
     """
-    cdef ThreadState state_ = <ThreadState>state
+    cdef _ThreadState state_ = <_ThreadState>state
     if what == PyTrace_LINE or what == PyTrace_RETURN:
         # Normally we'd need to DECREF the return from
         # `get_frame_code()`, but Cython does that for us