Handle registration marks within DICOM PHI de-identification

[dmoore247]

Is your feature request related to a problem? Please describe.
The solution currently over redacts. One notable example is redacting the 'L' or the Left

Describe the solution you'd like
Per our slack discussion, we can have a map of standard 1-3 letter registration marks. When evaluating the output of the VLM, or OCR, run it past the list of "reserved" words representing registration marks.

Describe alternatives you've considered
Tune selectivity.

Additional context

[dmoore247]

Registration marks in DICOM image pixel data can include:

"L" (Left) and "R" (Right), which denote anatomical orientation.
Other marks such as "A" (Anterior), "P" (Posterior), "S" (Superior), "I" (Inferior), depending on modality or protocol.
"U": Unpaired (used when the anatomical structure is not paired, such as the heart or spine).
"B": Both (used to indicate both left and right sides, such as both eyes or both knees imaged together).
Marks such as "AX" (axial), "COR" (coronal), "SAG" (sagittal), or protocol information indicating scan plane or acquisition parameters.
Text like "Test", "Calibration", or details about equipment settings.
Notations like "CH1" or "CH2" denoting which sensor channel produced the image
Calipers are commonly used in ultrasound to measure lesions or anatomical structures; typical burned-in calipers include sets of two or more perpendicular lines with numbers indicating distances (e.g., “12.4 mm”)

[dmoore247]

This spec section "C.7.6.1.1.1 Patient Orientation" is most complete section describing registration marks.
Assume single character registration marks fall in this list.