Add CancerVerse dataset support to the viewer backend

[aperson30]

What

Adds a second dataset — CancerVerse (22.8k CT-only cases, CV_%08d ids, CSV metadata) — alongside PanTS. Fully additive and dataset-dispatched: every PanTS path stays byte-for-byte unchanged, and CancerVerse is inert unless CANCERVERSE_PATH is set.

CancerVerse has no masks yet (no combined_labels, segmentations, or organ intensities), so every mask-dependent endpoint degrades gracefully to {"masks_available": false} instead of 404/crashing.

Changes (flask-server/)

• constants.py — CANCERVERSE_PATH, CANCERVERSE_LOWRES_PATH (default /home/visitor/cancerverse_lowres), DATASET_PREFIXES.
• api/utils.py — get_cancerverse_id, get_dataset_from_case_id, get_folder_id, get_case_nifti_paths; loads the CancerVerse CSV through the same _norm_cols into DF_CV (stays None if the path/CSV is absent — never raises); select_dataset_df() switches search on ?dataset=.
• api/api_blueprint.py — get-main-nifti serves CV CT + ?res=low from the CancerVerse paths; get-segmentations / get-specific-segmentations / get-report-data return {"masks_available": false} for CV; previews degrade gracefully; /api/search honours ?dataset=cancerverse|all.
• services/npz_processor.py, preprocess_meshes.py — get_cancerverse_id parallels.
• .env.example (new) + README — document the new env vars.

Notes / deviations from the spec

• utils.py was edited per explicit maintainer sign-off (normally off-limits). Kept surgical: one added entry to case_cols, the new helpers, and the DF_CV/select_dataset_df block — no existing search logic changed.
• Used a central dispatch helper (get_dataset_from_case_id + get_case_nifti_paths) rather than mechanically wrapping each scattered get_panTS_id call. PanTS branches are untouched; CV is a pure if. Same result, far lower regression risk.
• The spec said add two lines to .env — .env is gitignored, so those go in .env.example here and must be added to the server's real .env at deploy time (CANCERVERSE_PATH, CANCERVERSE_LOWRES_PATH).
• nifti_processor.py does not have a get_panTS_id copy (spec was mistaken), so nothing to add there.

Testing

• py_compile clean on all changed files
• Helper unit smoke: get_cancerverse_id('123')→CV_00000123, get_cancerverse_id('CV_00000123')→CV_00000123, dataset dispatch + masks_available correct, DF_CV gracefully None with no CV data
• 42/42 test_path_safety.py pass
• Server verification needed (no CancerVerse data locally): set the env vars, confirm ?dataset=cancerverse search returns CV rows, a CV case's CT loads, and mask endpoints return masks_available:false.

No gunicorn restart in this PR — that's the separate post-merge deploy step.

[aperson30]

Fixed the CSV path bug from review: CancerVerse_dataset_metadata.csv is now read from dirname(CANCERVERSE_PATH) (its real location, next to the image folder) rather than inside CANCERVERSE_PATH. Used os.path.dirname(os.path.normpath(...)) so a trailing slash on the env var is also handled, and corrected the .env.example + README notes. Pushed as 9856118.