Report CodegenError::Unsupported as a rustc diagnostic instead of panicking

[soumik15630m]

Summary

When Context::compile returns Err(CodegenError::Unsupported(_)),
src/base.rs currently falls through to the generic catch-all arm and
panics:

Err(err) => {
    panic!("Error while defining {name}: {err:?}", name = codegened_func.symbol_name);
}

This turns a recoverable, known-cause compilation error into an ICE.
The codebase already has a precedent for handling a specific
CodegenError variant gracefully — CodegenError::Verifier is caught
in its own arm just above and reported via early_dcx.early_fatal
instead of panicking. This PR adds a matching arm for
CodegenError::Unsupported.

Motivation

This gap was found while working on
bytecodealliance/wasmtime#13783, which changes x64 codegen to return
CodegenError::Unsupported instead of panicking when a function's
stack frame is too large to encode (see
#1656). Before that PR, Cranelift
never returned this error variant from this call site, so this gap in
cg_clif's own error handling was never exercised. With rust-lang/rustc_codegen_cranelift#13783's fix
applied, Context::compile now reaches this code path and previously
hit the generic panic!, producing an ICE with a stack trace instead
of a clean compiler error.

Testing

Built this against a local patch of bytecodealliance/wasmtime#13783
(via [patch.crates-io]) and rebuilt with y.sh build. Compiling the
repro from #1656
(let x = [0; 536870787_usize];) now produces:error: cranelift codegen error: stack frame size exceeds the 2GB limit supported by the x64 backend (see #1656)

instead of panicking on the Err at this call site.

Known follow-up (not fixed here)

With this change, the error is reported cleanly, but the worker thread
that calls early_fatal from inside this closure then triggers a
secondary panic in rustc_codegen_ssa ("expected abort due to worker
thread errors", back/write.rs:2107). This reproduces for the
existing Verifier arm too, using this same early_fatal pattern —
it's a pre-existing issue in how rustc's codegen coordinator handles a
worker thread calling early_fatal, tracked separately in #1664, and
out of scope for this PR.