Update cuda-oxide to Pliron 0.19.0

[nihalpasham]

Summary

Move both cuda-oxide compiler workspaces from the Pliron Git pin to the published 0.19.0 crates. Exact version pins make the dependency reproducible while retaining the configuration without the LLVM FFI bridge. This change targets main and applies to the existing compiler and examples.

Changes

• Pin pliron, pliron-derive, and pliron-llvm to =0.19.0, update combine, both lockfiles, and dependency source/license metadata.
• Implement Pliron's new branch operand-range and segment interfaces. Validate malformed metadata, including overflowing counts, before accessing operand ranges or allocating export state.
• Adapt printing, source locations, and attribute-map access. Fix function parsing to preserve exactly one body region and round-trip optional attributes.
• Add regressions for invalid branches, successor mutations, exporter error handling, and function parsing.

Testing

• just check passed from cuda-oxide/ on the pinned nightly: formatting, strict clippy, tests, guards, intrinsics, and rustdoc/doctests, including the separate backend workspace.
• cargo oxide run vecadd --arch sm_120 passed on RTX 5090: all 1,024 elements correct.
• cargo oxide run array_for_loop --arch sm_120 passed on RTX 5090: scalar and struct array iteration plus generated/mapped array helpers across 32 threads.
• Independent review covered the main-based migration, malformed branch metadata, parser invariants, and dependency resolution.
• New example: not applicable.

Checklist

• Commit signed and signed off.
• Existing SPDX headers preserved; no new source files.