Made reading of IPC dictionaries lazy

[jorgecarleitao]

This PR moves the reading of IPC dictionaries to when they are needed.

If a file has dictionaries, currently we read them to memory when reading the files' metadata. This causes metadata reading quite expensive if dictionaries exist.

This PR makes reading of dictionaries happen when the first IPC record batch is requested (via .next or poll for sync and async respectively), which allows users that just want to read the metadata to not have to read dictionaries.

The other advantage of this approach is that it allows projection push down to be applied to dictionaries, where we skip dictionaries that are not needed (follow-up PR).

[codecov]

Codecov Report

Merging #971 (224872f) into main (3a3e41b) will decrease coverage by 0.02%.
The diff coverage is 75.30%.

@@            Coverage Diff             @@
##             main     #971      +/-   ##
==========================================
- Coverage   71.53%   71.51%   -0.03%     
==========================================
  Files         355      355              
  Lines       19607    19626      +19     
==========================================
+ Hits        14026    14035       +9     
- Misses       5581     5591      +10     

Impacted Files	Coverage Δ
src/io/flight/mod.rs	0.00% <0.00%> (ø)
src/io/ipc/read/file_async.rs	57.02% <73.33%> (-3.85%)	⬇️
src/io/ipc/read/reader.rs	75.79% <80.00%> (-1.29%)	⬇️
src/compute/arithmetics/time.rs	26.60% <0.00%> (+0.91%)	⬆️
src/bitmap/utils/slice_iterator.rs	87.93% <0.00%> (+1.72%)	⬆️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 3a3e41b...224872f. Read the comment docs.