Fix NodeId->HirId lookups by carrying structural context (#1445)

c2rust-refactor would fail due to invalid AST node <-> HIR mappings.
This was caused by insufficient disambiguation due to all
c2rust-bitfield-derived expanded code sharing the same span with
call_site. This PR addresses the aforemented issue by threading
structural context through the AST and HIR walkers so NodeId-HirId
lookups stay precise even when spans collide.

- Added `child_slot`, `NodeContextKey`, and `StructuralContext` helpers
to record hierarchical positions for AST nodes; mirrored the same
traversal on the HIR side and stitched the maps together during `HirMap`
construction
- Collapsed wrapper nodes derived accessors and helper shims would erase
array/tuple shells on the expression side, leaving `Leaf` nodes that
conflicted with the HIR/type view where the same values remained wrapped
(e.g. `[u8; 3]` -> `Node([Leaf(u8)])`). We now detect those single-child
wrappers, peel just enough layers to expose the matching field list, and
otherwise panic with a targeted message if the arity still disagrees.

I tried going down another path with more-specific spans, such as using
quote_spanned!(field.field_name.span()), using fresh spans with
Span::def_site().located_at(old_span), and using attribute metadata
(with meta_name_value.lit.span(), but kept running into issues and so
abandoned that path.

Author: oinoom

c2rust-refactor/src/ast_manip/mod.rs

@@ -35,7 +35,9 @@ pub use self::load_modules::load_modules;
 pub use self::output_exprs::fold_output_exprs;
 pub use self::remove_paren::remove_paren;
 pub use self::seq_edit::{fold_blocks, fold_modules};
-pub use self::span_maps::{AstSpanMaps, NodeSpan, SpanNodeKind};
+pub use self::span_maps::{
+    child_slot, AstSpanMaps, NodeContextKey, NodeSpan, SpanNodeKind, StructuralContext,
+};
 pub use self::visit::Visit;
 pub use self::visit_node::{visit_nodes, visit_nodes_post, VisitNode};