feat: support negative matching in queries

[Chetnapadhi]

Summary

This PR implements support for negative matching in queries, enabling expressions such as "button not submit", "link without logout", and "input excluding email".

Changes

• Added query parsing to detect negative tokens with support for multiple triggers and multi-token negatives
• Integrated negative scoring in the lexical matcher using Jaccard similarity with synonym-aware penalties and hard exclusion thresholds
• Integrated negative scoring in the embedding matcher using cosine similarity-based penalties
• Centralized query parsing in the combined matcher to avoid redundant processing
• Ensured all scores are clamped within the range [0,1]
• Preserved existing behavior for positive-only queries

Tests

• Added table-driven tests for query parsing, including edge cases
• Added integration tests for negative matching across lexical, embedding, and combined matchers
• Covered cases for synonym-aware exclusion and negative-only queries
• Ensured no regression in existing positive query behavior

Validation

• go build ./... passed
• go test ./... passed
• go vet ./... passed
• No changes to public API
• No new dependencies introduced

Notes

• Negative matching supports multiple triggers such as "not", "without", "exclude", "except", and "no"
• Synonym expansion is applied to negative tokens for consistent matching behavior

Fixes #24

[Copilot]

Pull request overview

Adds negative matching support to the semantic element finder so queries can exclude/penalize elements using triggers like not, without, excluding, etc.

Changes:

• Introduces centralized query parsing into positive/negative token sets.
• Applies negative penalties/exclusions across lexical, embedding, and combined matchers (with score clamping).
• Adds tests and benchmarks covering negative matching, negative-only queries, and regressions.

Reviewed changes

Copilot reviewed 12 out of 12 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
semantic_test.go	Adds end-to-end tests for negative matching behavior in the public package surface.
internal/types/types.go	Adds ParsedQuery type to represent positive vs negative tokens.
internal/engine/query_parser.go	Implements query parsing with negative triggers.
internal/engine/query_parser_test.go	Adds table-driven tests for query parsing edge cases.
internal/engine/lexical.go	Integrates negative parsing and penalties into lexical matching.
internal/engine/lexical_test.go	Adds lexical unit tests for negative penalization, synonyms, and empty queries.
internal/engine/embedding.go	Integrates negative parsing and cosine-similarity penalties into embedding matching.
internal/engine/embedding_test.go	Adds embedding unit tests for negative penalization, negative-only, and empty queries.
internal/engine/combined.go	Centralizes parsing and fuses results with clamped combined scores.
internal/engine/combined_test.go	Adds combined-matcher coverage for negative queries, regressions, and clamping.
internal/engine/benchmark_test.go	Fixes element ref generation and adds benchmark scenarios for negative queries.
README.md	Documents negative query syntax, examples, and expected behavior.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[luigi-agosti]

@Chetnapadhi thanks, I'm going to do a follow up on this to fix a couple of things