errors: remaining span-less errors — frontmatter name collisions and other sites without byte offsets

[dean0x]

v0.4.0 (PR #196) added spans to errors wherever byte offsets were available. However, some error sites still lack spans: frontmatter key collisions (no YAML byte offset in scope) and other boundary errors.

Current situation: Most error classes now carry spans, but frontmatter-origin errors remain spanless. Additionally, the theoretical name_collision_at class is currently guarded by an unconditional OOB degradation in at().

Open question: Can we thread byte offsets through the YAML frontmatter parser to add spans to collision diagnostics? Should we audit and close the remaining spanless error classes?

Deferred from the v0.4.0 dogfooding remediation (PR #196) — flagged during the 2026-07-17 dogfooding campaign; deliberately not implemented there pending design discussion.

[dean0x]

Housekeeping: the cross-reference to this issue from #304 was a typo for #303 (the apply_fixes deprecation PR) — #304's body has since been corrected, so this reference is unrelated to this issue and can be ignored.