US7ASCII database: charset conversion corrupts multi-byte pass-through data

[rophy]

Version: dabeb4b1 (master)

Description

Legacy Oracle databases using US7ASCII often store multi-byte characters (BIG5, GB2312) as raw bytes — a common practice known as "pass-through". OLR reads nls-character-set: US7ASCII from the schema and applies charset conversion, stripping the high bit from every byte >= 0x80, which destroys the original data.

A config option to skip charset conversion and emit raw bytes as-is would solve this.

Steps to reproduce

1. Oracle XE 21c with NLS_CHARACTERSET = US7ASCII
2. Insert Big5-encoded Chinese characters as raw bytes:

CREATE TABLE TEST_MULTIBYTE (id NUMBER PRIMARY KEY, name VARCHAR2(200));
ALTER TABLE TEST_MULTIBYTE ADD SUPPLEMENTAL LOG DATA (ALL) COLUMNS;
-- Big5: 台北 = A578 A55F
INSERT INTO TEST_MULTIBYTE VALUES (1, UTL_RAW.CAST_TO_VARCHAR2(HEXTORAW('A578A55F')));
COMMIT;

3. Capture redo logs and run OLR in batch mode

Expected result

Raw bytes preserved in JSON output: "NAME" contains bytes A5 78 A5 5F.

Actual result

OLR strips the high bit (& 0x7F) from every byte >= 0x80:

Big5 input: A5 78 A5 5F  (台北)
OLR output: 25 78 25 5F  (%x%_)

{"after":{"ID":1,"NAME":"%x%_"}}

The original Big5 data is unrecoverable from OLR's output.