fix(migrate): under() panics when a trigger length splits a multi-byte char

[charlesdove977]

@ChristopherKahler heads up, this one takes down the SessionStart hook for any workspace that has a markdown file with a multi-byte char in its path.

Bug

base hook session-start panics on 0.15.1 and 0.15.2 (and current main):

thread 'base-main' panicked at src/migrate.rs:704:15:
end byte index 49 is not a char boundary; it is inside '→' (bytes 48..51 of string)

under() byte-slices rel[..dir.len()]. When a path like graphify-out/obsidian/Auto-reply → engagement → algorithm boost.md (graphify's Obsidian export names notes this way) has a multi-byte char straddling the byte offset equal to a trigger's length, the slice panics and the whole hook exits 101. Claude Code shows it as SessionStart:startup hook error every session.

Fix

rel.get(..dir.len()).is_some_and(|head| head.eq_ignore_ascii_case(dir)). get returns None off a char boundary; a split char means the prefix can't match anyway, so it's a miss instead of a crash. Behavior on every valid boundary is unchanged.

Verification

• New assert !under("notes/a → b.md", "notes/a x") in under_is_a_path_test_not_a_string_prefix: panics at migrate.rs:704 without the fix, passes with it.
• Release build of this branch runs base hook session-start in the affected workspace with exit 0.
• cargo test: 556 pass. graph::tests::purge_apply_deletes_unread_spares_recent_and_snapshots and purge_dry_run_selects_by_recency_and_writes_nothing fail identically on unmodified main locally, so they're unrelated to this change.

🤖 Generated with Claude Code