perf(image-redactor): index analyzer bboxes by position in get_pii_bboxes

[SiddhaBasu]

Change Description

ImageAnalyzerEngine.get_pii_bboxes rescanned the entire analyzer_bboxes list for every ocr_bbox to find a positional match (left/top/width/height equality), an O(len(ocr_bboxes) × len(analyzer_bboxes)) scan. This builds a dict keyed by (left, top, width, height) from analyzer_bboxes once up front, so each ocr_bbox is resolved with a single O(1) dict lookup instead — O(n) + O(m) total.

No behavior change. The original loop used break on the first matching analyzer_bbox, so when multiple analyzer bboxes shared a position, the first one in list order always won. A naive dict[key] = value build would flip that to last-wins, so this uses dict.setdefault(key, analyzer_bbox) to keep identical first-match semantics. Verified with:

• A new regression test asserting the first of two same-position analyzer bboxes is the one kept.
• A new equivalence test comparing this implementation against a preserved copy of the original O(n·m) logic on 500 bboxes — output is identical.

Scope note: get_pii_bboxes is called from ImagePiiVerifyEngine.verify() and DicomImagePiiVerifyEngine — the verification/visualization tooling that draws PII/non-PII bounding boxes on an image for inspection. The core ImageRedactorEngine.redact() path does not call this method, so this improves the verify/QA tooling, not the redaction pipeline itself.

Issue reference

N/A — found while auditing presidio-image-redactor for nested-loop hotspots; no existing issue tracked this.

Testing

• pytest tests/test_image_analyzer_engine.py: 5 existing get_pii_bboxes tests pass, plus 3 new tests added.
• pytest tests/test_dicom_image_pii_verify_engine.py (a direct caller of this method): unchanged pass/fail counts before and after this change.
• Full presidio-image-redactor suite: identical failure/error counts before and after this change (confirmed via git stash); only difference is the added tests passing.
• ruff check / ruff format --check on the changed source file: clean.
• Could not produce a numeric coverage percentage locally — coverage.py instrumentation crashes on an unrelated torch/thinc import ordering issue in my local environment, reproducible on unmodified main as well. Every branch touched by this diff (dict build, match found, no match, duplicate-position tie-break) is exercised by the test suite.

Checklist

• I have reviewed the contribution guidelines
• I agree to follow this project's Code of Conduct
• I confirm that I have the right to submit this contribution and that it does not knowingly contain proprietary or confidential code
• My code includes unit tests
• All unit tests and lint checks pass locally
• My PR contains documentation updates / additions if required (none required — no public API/behavior change)