Use UstrSet for TypeInfo.properties_visited and TypeInfo.aliases

[kennethloeffler]

For better rbx_binary serialization performance, this PR changes the type of TypeInfo.properties_visited from HashSet<(Ustr, VariantType)> to UstrSet, and changes the type of TypeInfo.aliases from BTreeSet<Ustr> to UstrSet. These changes are sound because:

• The second tuple field in TypeInfo.properties_visited entries is not used for anything outside of the collection, and it's impossible for an instance to contain multiple properties that have the same name, but different types.
• The order of TypeInfo.aliases is unimportant.

On my machine, serialization benchmark results compared to latest master are as follows:

100 Folders/Serialize   time:   [20.219 µs 20.256 µs 20.298 µs]                                   
                        thrpt:  [37.165 MiB/s 37.242 MiB/s 37.309 MiB/s]
                 change:
                        time:   [-2.1144% -1.9337% -1.7269%] (p = 0.00 < 0.05)
                        thrpt:  [+1.7573% +1.9719% +2.1600%]
                        Performance has improved.

100 Deep Folders/Serialize                                                                             
                        time:   [22.818 µs 22.837 µs 22.858 µs]
                        thrpt:  [33.001 MiB/s 33.032 MiB/s 33.060 MiB/s]
                 change:
                        time:   [-2.5076% -2.3625% -2.2207%] (p = 0.00 < 0.05)
                        thrpt:  [+2.2711% +2.4196% +2.5721%]
                        Performance has improved.

100 100-line ModuleScripts/Serialize                                                                            
                        time:   [76.197 µs 76.256 µs 76.313 µs]
                        thrpt:  [48.338 MiB/s 48.374 MiB/s 48.412 MiB/s]
                 change:
                        time:   [+0.9836% +1.5364% +2.1002%] (p = 0.00 < 0.05)
                        thrpt:  [-2.0570% -1.5131% -0.9740%]
                        Change within noise threshold.

1,000 Parts/Serialize   time:   [1.3277 ms 1.3290 ms 1.3303 ms]                                   
                        thrpt:  [3.1464 MiB/s 3.1495 MiB/s 3.1526 MiB/s]
                 change:
                        time:   [-6.6982% -6.5511% -6.4099%] (p = 0.00 < 0.05)
                        thrpt:  [+6.8490% +7.0104% +7.1791%]
                        Performance has improved.

[Dekkonot]

I can't validate your performance claims because the benchmark results are honestly all over the place for me (the cost of being on Windows, I guess), but it doesn't seem to hurt anything so really all it can do is improve performance. LGTM.

[kennethloeffler]

To help reduce noise, it's generally good to have as few other things running on your system as possible. 1,000 parts is definitely more prone to problematic pollution than 10,000 parts too, so if I do any more performance work in the future, I'll make sure to add some chunkier (and more diverse) benchmarks