fix: resolve compiler panic with large exponent constant rationals

[ParthSinghPS]

Fixes #1881

Description
The compiler was occasionally panicking with Option::unwrap() on a None value in eval_const_rational when a constant with a large scientific-notation exponent (e.g. 0e1000) was used in a rational type context (such as array bounds).

When the parser builds the AST, number_literal() correctly rejects an exponent as too large, leaving the constant variable's initializer as None. However, later evaluations simply used .unwrap() on the initializer. This PR replaces those calls with proper if let Some error handling, returning a Diagnostic instead.

Changes

• Added missing initializer checks in sema/eval.rs for ConstantVariable evaluation branches.
• Added a polkadot contract test case large_exponent_const_rational.sol to ensure the compilation errors gracefully rather than crashing.