Preserve original anchor names through propagation

[anthrotype]

After #1832 (propagate anchors in IR), fonts with non-standard caret anchor naming lost caret positions. For example, ComforterBrush-Pro.glyphs defines two caret anchors on "ffi":

• caret_1 at x=242
• caret_1_ at x=588

AnchorKind::new("caret_1_") parses the suffix "1_", fails parse::<usize>(), and falls back to Caret(1), i.e. the same variant as caret_1. When to_name() method converts both back to strings during anchor propagation, both become "caret_1" and one overwrites the other in the IndexMap.

Before #1832 this didn't matter because fontbe consumed Vec<Anchor> directly without round-tripping through names. After the merge, propagation serializes anchors via AnchorKind::to_name() at

https://github.com/googlefonts/fontc/blob/ade0a9bc/fontir/src/propagate_anchors.rs#L87

... introducing the lossy conversion.

(fontmake/ufo2ft doesn't have this problem because it never parses caret indices, it just checks name.startswith("caret_") and collects positions.. apparently nobody cares about caret indices except me)

The fix is to add original_name: SmolStr field to fontir::ir::Anchor to preserve the original source name and use that instead of anchor.kind.to_name() when feeding anchors into propagation.

The existing to_name() is kept as a public method but documented as lossy.

The first commit adds the regression test that asserts both caret_1 and caret_1_ survive propagation (deliberately fails without the fix). It is followed by the actual fix.

[anthrotype]

https://github.com/googlefonts/fontc/actions/runs/21998353880/job/63564211580?pr=1890#step:5:246

thread 'propagate_anchors::tests::caret_anchor_names_preserved_through_propagation' (4227) panicked at fontir/src/propagate_anchors.rs:2062:9:
assertion `left == right` failed
  left: ["caret_1", "top_1", "top_2", "top_3"]
 right: ["caret_1", "caret_1_", "top_1", "top_2", "top_3"]