State loading optimizations

[booky10]

See individual commits for more details, but essentially:

• Reduced retained memory after states have been loaded (array instead of map and no longer allocates some ints)
• Reduced allocations during state loading (e.g. sequential nbt reading now only uses one entry object)
• Buffer gzip input stream (buffer was previously placed incorrectly)
• Always compare/hash state types + nbt types by object identity

On my device, all states now load in about 1.3s-1.5s instead of 2.1s-2.3s like before

---

Some async profiler reports:

• alloc: https://files.booky.dev/public/26-06-26_asprof_packetevents_alloc.html
• cpu: https://files.booky.dev/public/26-06-26_asprof_packetevents_cpu.html

If someone sees/finds anything in there which could be further optimized, please let me know!

---

Initially marked as a draft as I still need to test + check if everything is behaving correctly

(thanks to MissingReports on Discord for suggesting to take a look at this!)