Implement eager mode caching

[pranavm-nvidia]

We should implement a caching mechanism for eager mode so that we don't need to compile each time. We should make it so that every constant (i.e. produced by a Storage op) becomes an input. That way, we only need to compare the text IRs (i.e. the graph structure) to check for cache hits.

[farazkh80]

Per discussion we should also inlcude

• cache mode enable/disable using environment variables
• size for cache memory (i.e. 2GB)
• make sure the cache key string includes all ops vars and kwargs

[pranavm-nvidia]

We can add the environment variable in config.py