perf: speed up fsst decompression

[broccoliSpicy]

before:

after:

to reproduce:
cargo run --release --example benchmark
in rust/lance-encoding/compression-algo/fsst

machine info:
11th Gen Intel(R) Core(TM) i7-1165G7 @ 2.80GHz
Linux 192 5.10.0-28-amd64 #1 SMP Debian 5.10.209-2 (2024-01-31) x86_64 GNU/Linux

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 98.22485% with 3 lines in your changes missing coverage. Please review.

Project coverage is 79.39%. Comparing base (02294a1) to head (2a40120).
Report is 1 commits behind head on main.

Files	Patch %	Lines
...t/lance-encoding/compression-algo/fsst/src/fsst.rs	98.22%	3 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2626      +/-   ##
==========================================
+ Coverage   79.35%   79.39%   +0.04%     
==========================================
  Files         213      213              
  Lines       62520    62706     +186     
  Branches    62520    62706     +186     
==========================================
+ Hits        49610    49784     +174     
- Misses       9996    10005       +9     
- Partials     2914     2917       +3     

Flag	Coverage Δ
unittests	79.39% <98.22%> (+0.04%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[broccoliSpicy]

end-to-end test:

to reproduce:

1. disable dictionary encoding
2. download dataset
   wget https://huggingface.co/datasets/HuggingFaceFW/fineweb/resolve/main/data/CC-MAIN-2013-20/000_00000.parquet\?download\=true
3. run script(set LANCE_USE_FSST environmental variable to enable fsst and unset it to disable)

from lance.file import LanceFileReader, LanceFileWriter
import pyarrow.parquet as pq

parquet_file_path = "/home/x/data.parquet"
data = pq.read_table(parquet_file_path)
lance_file_path = '/home/x/lance-experiments/fineweb/output.lance'
with LanceFileWriter(lance_file_path) as writer:
  writer.write_batch(data)

import datetime
import pyarrow.parquet as pq
from lance.file import LanceFileReader

start = datetime.datetime.now()
tab = pq.read_table(parquet_file_path)
end = datetime.datetime.now()
elapsed = (end - start).total_seconds()
print(f"Parquet elapsed: {elapsed}s")

start = datetime.datetime.now()
tab2 = LanceFileReader(lance_file_path).read_all().to_table()
end = datetime.datetime.now()
elapsed = (end - start).total_seconds()

import os  


lance_file_size = os.path.getsize(lance_file_path)
lance_file_size_mib = lance_file_size // 1048576
parquet_file_size = os.path.getsize(parquet_file_path)
parquet_file_size_mib = parquet_file_size // 1048576

if os.getenv("LANCE_USE_FSST") is not None:

  print(f"Parquet file size(fsst): {parquet_file_size_mib} Mbytes")
  print(f"Lance file size(fsst): {lance_file_size_mib} Mbytes")
  print(f"Lv2(fsst) elapsed: {elapsed}s")
else:
  print(f"Parquet file size(fsst): {parquet_file_size_mib} Mbytes")
  print(f"Lance file size(no fsst): {lance_file_size_mib} Mbytes")
  print(f"Lv2(no fsst) elapsed: {elapsed}s")

assert tab == tab2

print("Tables are equal")

[westonpace]

Sorry for the delay in getting back to this. I don't yet fully understand FSST and so I don't understand exactly this change. However, the tests pass and FSST is still guarded by env variable so lets merge this and I will try and set some time aside next week to really dig through in more detail.

[broccoliSpicy]

ha, sorry that I didn't explain this PR, it is basically a rust translation from the original c++ implementation of decompression.
and uses rust's ptr::write_unaligned as a performant way to do store