* Add children sorting to snapshot_reconciler
* Update snapshot tests to include stable children order
* Bump dependencies, which should make this PR work
* Changes project-related structures to use `BTreeMap` instead of `HashMap` for children to aid determiniusm
* Changes imfs-related structures to have total ordering and use `BTreeSet` instead of `HashSet`
* Upgrades dependencies to `bx_dom_weak`1.2.0 and rbx_xml 0.5.0 to aid in more determinism stuff
* Re-exposes the `RbxSession`'s root project via `root_project()`
* Implements `Default` for a couple things
* Tweaks visualization code to support visualizing trees not attached to an `RbxSession`
* Adds an ID-invariant comparison method for `rbx_tree` relying on previous determinism changes
* Adds a (disabled) test to start finding issues in the reconciler with regards to communicativity of snapshot application
* Adds a snapshot testing system that operates on `RbxTree` and associated metadata, which are committed in this change
* Add 'plugins' field to project and add rlua
* Scaffold out new SnapshotContext type (again) with plugin state
* Almost functional snapshot system with rlua proof-of-concept
* Gate plugin config on 'plugins-enabled' feature, tell Travis to test all features
* Guard remaining plugin setup code behind feature
* Bump minimum version to 1.33, should've caught this before
* Whoops, latest Rust is 1.32, not 1.33
* Do the nested partition thing
* Tidy up touched code
* Add nested partition test project, not fully functional
* Clean up variable names, move path_metadata mutation strictly into snapshot_reconciler
* Remove path_metadata, snapshotting is now pure
* Factor out snapshot metadata storage to fix a missing case
* Pull instance_name out of per_path_metadata, closer to what we need
* Refactor to make metadata make more sense, part one
* All appears to be well
* Cull 'metadata_per_path' in favor of 'instances_per_path'
* Remove SnapshotContext
* InstanceMetadata -> PublicInstanceMetadata in web module
* Build in snapshot testing system for testing... snapshots?
* Remove pretty_assertions to see if it fixes a snapshot comparison bug
* Reintroduce pretty assertions, it's not the cause of inequality
* Fix snapshot tests with custom relative path serializer
