Story: Carry six-digit microseconds in every timestamp the platform writes

Every timestamp the platform writes carries six digits of microseconds, and the parser reads them back, so a time_point survives a round trip through the database or the wire to the microsecond.

• to_iso8601_utc writes six fractional digits and parse_utc_timestamp reads one to six back.
• A time_point survives to_iso8601_utc then from_iso8601_utc to the microsecond.
• The market observation test asserts the round trip and passes.

• The writer and the reader change together. Reason: a writer that states a fraction the reader ignores looks fixed until the value is read back.
• Six digits, always. Reason: PostgreSQL states none at zero microseconds, so the reader accepts an absent fraction, but a writer that varies its width makes the wire form unstable.
• The market ingest assertion is tightened back to the millisecond. Reason: main had answered the same defect by weakening it to a whole second and recording the truncation as intended; the fix makes the millisecond true.
• The platform and wire layers change; the entity layer does not. Reason: db_timestamp is rfl::Timestamp, which wraps a std::tm with no sub-second field, so no format string can give it a fraction.

• Replacing db_timestamp so the entity layer's audit columns and its mappers also carry microseconds. That is a change to every entity and mapper, not a format string, and it needs its own story.