fix(core): accept split tokenizer files for Sherpa Qwen

[H-Chris233]

Summary

• accept both tokenizer/tokenizer.json and the supported tokenizer/vocab.json + tokenizer/merges.txt layout for Sherpa Qwen archives
• reuse the same completeness rule for installed-state checks, legacy migration, and post-extraction validation
• add regression coverage for Beta.1 upgrades and fresh archive extraction

Root cause

The Core ModelStore migration narrowed the previous Sherpa Qwen tokenizer contract to tokenizer/tokenizer.json only. Existing models and the release archive can use the split BPE files, so Beta.1 rejected valid models before runtime loading.

Tests

• cargo test -p openless-core (795 passed, 1 ignored)
• cargo test -p openless-core model_store::tests:: (23 passed)
• rustfmt --edition 2021 --check crates/openless-core/src/model_store.rs
• git diff --check upstream/beta...HEAD