Refactor and update parsing

[moofMonkey]

Fixes #7
Fixes #8

• Fix docker build
• Fix stringtables parsing
• Fix deprecation warnings
• Add support for stringtables using varint_bitcounts
• Fix fp_PushN
• Clean up and optimize fieldpath code
• Implement resource manifest parsing to avoid embedding resources trie in butterfly binary/sources
• Rework flattened serializers
  • Handle all arrays as if they're dynamic
  • Move property#type to fs_info
  • Avoid storing property paths as strings and introduce new property getter API to reduce RAM usage
  • Construct entities layout inside flattened_serializer instead of using std::vector/std::unordered_map to reduce RAM and CPU usage
• Fix m_fGameTime logic - m_fGameTime was removed in favor of tick-based time calculation
• Implement proper entities baselines support
• Implement stringtables callbacks
• Implement entities props callbacks nor specify changed paths in on_entity callback?
• Eliminate BUTTERFLY_THREADSAFE
  • Eliminate global object pools
  • Replace static std::vector with llvm::SmallVector in entity::parse
• Discuss and reimplement JS/Python APIs using event-driven approach
• Discuss whether we want to store resource paths as strings or provide API for resource path resolution
• Discuss clang-format
  Would it be fine if I add Google/LLVM-styled clang-format?
  If so, is SpacesInParentheses: true worth preserving in butterfly code?
• Fix Windows builds
• Run unit tests against matches in different game patches?
• Add support for GitHub CI?
• Store statically sized arrays without additional allocations?
  This might introduce problems if array size is parsed incorrectly, but could theoretically reduce RAM/CPU usage.

Known problems:

• Possible crash on exit in case of global butterfly::parser due to global object pools heap use-after-free in entity::~entity
  This affects 03-props, 06-events and 07-combatlog
• entity_classes class_id should either assert or return UINT32_MAX on lookup failure

Next PRs TODO:

1. Proper VPK API
2. KV parsing API

[moofMonkey]

Even though current DEFAULT_MAX_PATH_DEPTH is 10, no parsers currently handle this.
Having only 7 path elements gives us perfect 16 byte fieldpath size/alignment.