Startup time improvements

general guidelines

• Only make changes supported by profiling
• Qualify type-related imports under quicken.lib._typing.MYPY_CHECK_RUNNING and use from __future__ import annotations (may be revisited later depending on whether we want Python 3.6 support)
• Distinguish module import cost due to filesystem and actual execution with strace -c python -X importtime -c 'import <module>'.
• When checking cost of module import in isolation make sure to take into account modules that will already be imported (e.g. socket, os, re, maybe importlib).
• Use import stubbing (see quicken.lib._imports) as a last resort - prefer to remove dependencies entirely
• Nit-picky time savings only really apply to library code - tests can use whatever they need to

Some nice things that unfortunately don't seem worth it for our use case (but this should be checked later):

• dataclasses - module execution time is longer than with basic class definitions
• pathlib - adds a few ms due to urllib.parse and ntpath