WIP: trace func threading support, special-casing

line_profiler/_line_profiler.pyx
    TraceCallback
        Renamed from `CTraceState`
    alloc_callback(), free_callback()
        New functions for allocating `TraceCallback` objects
    c_trace_state_has_hook()
        Merged into `call_cached_callback()`
    fetch_callback()
        - Refactored and renamed from `fetch_c_trace_state()`
        - No longer `malloc()`-ing and returning a pointer, instead
          assigning to the fields of the pointer passed as the argument
    restore_callback()
        - Refactored and renamed from `restore_c_trace_state()`
        - No longer `free()`-ing the supplied pointer, instead just
          assigning `NULL` to its fields
    call_callback()
        - Merger of `call_c_trace_state_hook[_safe]()`
        - Now staying entirely in C-land (where possible), avoiding
          calling the `PyObject` attribute-access functions
        - Refactored heap-allocated swap `TraceCallback` objects into
          stack-allocated ones
        - Added check for disabling of the `sys` trace function -- if
          that happens, profiling continues (as with before) but the
          wrapped/cached callback is no longer called
        - Added check for disabling of frame-local line events -- if
          that happens, `.f_trace_lines` is restored (as with before)
          but the frame-local trace callable (`.f_trace`) is wrapped and
          shielded from line events
    get_optional_attr(), set_optional_attr()
        Removed now-unused methods
    disable_line_events()
        New Python-level wrapper for shielding a frame-local trace
        callable from line events
    LineProfiler.wrap_trace
        Cythonized definition
    LineProfiler.{dis,en}able(), python_trace_callback()
        Now handling per-thread caching of the `sys` callback state

Author: TTsangSC