[Bug] Overly aggressive secret redaction corrupts a11y snapshot labels and attributes #1106

Summary

The _redactSecrets function indiscriminately redacts any occurrence of a secret's value across the entire a11y snapshot. This leads to the corruption of non-input text, such as labels and attributes, whenever they contain a substring that matches a secret.

Example

Configure secrets such as:

• PASSWORD=ssw (short value)
• or PASSWORD=e2

Generate an a11y snapshot containing a textbox labeled:

• textbox "Password" [ref=e24]:

Observe the redaction result:

• With PASSWORD=ssw:
  textbox "Pa<secret>PASSWORD</secret>ord" [ref=e24]:
• With PASSWORD=e2:
  textbox "Password" [ref=<secret>PASSWORD</secret>4]:
  These substitutions happen even though the secret appears only as a substring of unrelated text/attributes.

Expected behavior

• Redact only the concrete input values that were filled from secrets (e.g., textbox … textInput: "•••"), or otherwise ensure redaction respects value boundaries and context so that labels/attributes are unaffected.

Impact

Short secrets (e.g., CVV codes) are prone to substring collisions, leading to the widespread masking of unrelated text and the corruption of a11y tree data.

Question

Can we scope redaction exclusively to input values (e.g., the textInput portion) instead of the entire a11y snapshot?