Use the upodesh crate for dictionary suggestion search

[mominul]

upodesh uses an approach based on the Finite State Transducer (FST) data structure which is substantially faster than the Regular Expression based approach. This approach is inspired by the Go project libavrophonetic of Mehdi Hasan Khan which used Trie data structure.

Benchmarks

upodesh is significantly faster than the previously used heavily optimized regex-based search approach in riti. Based on recent benchmarks, it is approximately ~21× to ~58× faster, depending on the input. This demonstrates a substantial performance gain over regex, especially in cases where large patterns previously caused bottlenecks.

📊 Summary of the Benchmark

This benchmark was performed on a Apple MacBook Air M1:

Word	upodesh Time	regex Time	Speedup
a	~3.341 µs	~194.34 µs	~58× faster
arO	~11.840 µs	~246.53 µs	~20.8× faster
bistari	~9.734 µs	~353.74 µs	~36.3× faster

cc @mugli @gulshan