fix(soroban): reject unsupported ABI types before codegen

[mohamedbasuony]

Summary

Fixes the compiler-side Soroban diagnostics covered by #1897.

Unsupported Soroban paths could previously reach codegen, Soroban encoder/decoder logic, spec emission, or LLVM and fail with internal panics, todo!(), unimplemented!(), Inkwell errors, or LLVM assertions. This PR adds Soroban codegen-side validation so those user-triggerable cases are reported as normal Solang diagnostics before the unsafe paths are reached.

What changed

• Move unsupported Soroban ABI validation out of sema and into the Soroban codegen path.
• Validate unsupported public/external Soroban function parameters and return values before dispatch wrapper generation.
• Validate unsupported public accessor return values before generated accessor code can reach unsupported Soroban lowering.
• Validate unsupported Soroban event argument types before event encoding.
• Guard multiple external return values and unsupported array/struct/string/bytes cases that could otherwise reach Soroban spec, encoder/decoder, Inkwell, or LLVM failures.
• Add precise diagnostics for internal string-helper patterns that could otherwise trigger LLVM/Inkwell failures.
• Keep supported Soroban behavior unchanged, including supported event emission and public string accessors.
• Target-gate the Soroban validation and early returns so non-Soroban diagnostic accumulation is unchanged.

Notes

• This does not catch panics globally.
• This does not wrap compiler execution.
• This does not suppress LLVM assertions.
• Unsupported Soroban cases are rejected before the relevant codegen/emit paths are reached.
• Non-Soroban targets are not affected by these Soroban diagnostics.

Verification

Ran locally:

• cargo fmt --all -- --check
• cargo build --workspace
• cargo test --test soroban
• cargo test --test contract solana_contracts -- --nocapture
• cargo clippy --workspace
• cargo test --workspace

Manual checks:

• dynamic_bytes_arg.sol: emits a normal diagnostic; old Soroban decoder panic is gone.
• string_return.sol: emits a normal diagnostic; old LLVM abort is gone.
• event_panic.sol: still compiles on this checkout.
• examples/soroban/token.sol: still compiles with existing warnings.
• Unsupported event argument stress cases emit diagnostics instead of panicking.
• Internal string-helper stress cases emit diagnostics instead of reaching LLVM/Inkwell failures.
• Public string accessor, internal bytes helper, and non-Soroban bytes/string controls remain unaffected.

Fixes #1897.

[salaheldinsoliman]

This change shouldn't be in sema. Instead, it should be handled in codegen, where target specific encoding/decoding logic is implemented .
Moreover, I expected to see:
1- panic!s without error messages addressed
2- Areas where LLVM panics could occur safely guarded by human readbale messages.

Changes should be mostly in emit and codegen

[salaheldinsoliman]

Left a comment

[mohamedbasuony]

Thanks, I see your point. I put the checks in sema as target validation, but I agree that this is too early/general for Soroban-specific ABI handling. The PR will be updated so that it handles unsupported Soroban ABI cases in the emit/codegen path, closer to the encoding/decoding logic.