fix(oracle): BoxLang reads a cf_sql_timestamp back as an unreadable oracle.sql.TIMESTAMP #3714

Found by the compat matrix once #3710 (#3708) let the BoxLang + Oracle leg run past fixture population. Before #3710 this leg ran zero specs, so this could not be seen.

Leg: boxlang + oracle, dispatch 36341832355, on #3710's head 8b2a9f814 (current with develop). Totals: 5815 specs, 5786 pass, 0 fail, 5 error, 24 skip.

Error: TimestampRoundTripSpec > "cf_sql_timestamp round-trip (#3302)" > "reads back as a CFML date or as epoch milliseconds":

Can't cast oracle.sql.TIMESTAMP to a string.
  at vendor/wheels/tests/specs/database/TimestampRoundTripSpec.cfc:74

There are two defects:

1. On BoxLang, Oracle returns a cf_sql_timestamp column as an oracle.sql.TIMESTAMP object. $normalizeDbTimestamp() does not turn that into a CFML date, so shape is "neither". Every framework path that reads a stored timestamp back is affected.
2. The spec's failure message concatenates the raw readBack value into a string. With a Java object, building the diagnostic throws, and that hides the real assertion. A diagnostic must never crash on a raw object.

Fix: teach the normalizer to read oracle.sql.TIMESTAMP (through its exported JDBC conversion, as a java.sql.Timestamp), and make the spec message print a safe representation of non-simple values.

Blocks 4.1.1: the release gate is every test passing on every matrix leg.