refactor: refuse to run transforms on crates that do not compile

[thedataking]

c2rust-refactor runs the rustc analysis passes before applying a transform, but discarded the result (let _result = tcx.analysis(())). When the input crate has type errors, the transform then operates on incomplete typeck tables: it may silently emit rewrites derived from bogus types, or panic on an internal invariant far from the actual problem. Because each transform in a sequence consumes the previous one's output, one transform that produces an invalid crate poisons every transform that runs after it.

This surfaced while adding libgit2 to the integration testsuite (#1861). libgit2 compiles its test sources with GIT_DEPRECATE_HARD but its library sources without it, so several public structs legitimately have two incompatible declarations within one crate (a member that is a function pointer in one view is a reserved void pointer in the other). The reorganize_definitions transform, which merges duplicate declarations by name, produced an ill-typed crate from that input, but the failure CI reported was a "mismatched key trees" panic in remove_literal_suffixes, three transforms later, which took considerable effort to trace back to the transform actually at fault.

Check the analysis result and exit with an error instead, so a transform sequence stops at the transform that broke the crate and rustc's diagnostics point at the real breakage.

[thedataking]

Withdrawing this: the refactor tool intentionally supports running commands on crates that do not compile (e.g. test_abstract feeds a crate with expected E0133 errors back into the tool, and the illtyped machinery exists to repair broken intermediate states), so an unconditional abort breaks that contract — as CI on #1864 demonstrated. The libgit2 failure that motivated this is addressed at the source in #1861 (transpiling with -UGIT_DEPRECATE_HARD), and the underlying reorganize_definitions limitation is tracked in #1899.