perf(python): Pre-allocate size for the dictionary

[penguin-wwy]

What does this PR do?

Pre-allocate memory for the dictionary based on the data size to avoid resizing and improve deserialization performance.

Related issues

Does this PR introduce any user-facing change?

• Does this PR introduce any public API change?
• Does this PR introduce any binary protocol compatibility change?

Benchmark

# python format
fury_large_dict: Mean +- std dev: [dict_base] 548 us +- 33 us -> [dict_resize] 531 us +- 33 us: 1.03x faster

# xlang format
fury_large_dict: Mean +- std dev: [dict_xlang_base] 550 us +- 39 us -> [dict_xlang_resize] 527 us +- 35 us: 1.05x faster

[chaokunyang]

LGTM