Redact credentials without mangling surrounding text

[openai0229]

What

redact() replaced the password and username anywhere they appeared as plain substrings. Testing the Oracle failure path showed the consequence: with the password oracle, the driver's documentation URL came out as https://docs.[redacted].com/error-help/db/ora-17002/, and a short username such as app would rewrite every application in a message.

A secret is now replaced only when it stands alone: the character before and after it must be neither alphanumeric nor a dot. Credentials echoed in connection text (jdbc:mysql://app:s3cret@db:3306/app) are still redacted, hostnames and ordinary words are left intact.

Verification

cargo test -p sqlx-protocol adds two unit tests:

redact("https://docs.oracle.com/error-help/db/ora-17002/", user=app, password=oracle)
  → unchanged
redact("the application could not start", …)                     → unchanged
redact("password authentication failed for user \"app\" with s3cret", …)
  → password authentication failed for user \"[redacted]\" with [redacted]
redact("jdbc:mysql://app:s3cret@db:3306/app", …)
  → jdbc:mysql://[redacted]:[redacted]@db:3306/[redacted]

cargo fmt --all -- --check and cargo clippy --workspace --all-targets --locked -- -D warnings passed.

Risks and limits

• A secret glued to a word or a longer hostname (for example a password that is a substring of the database name) is no longer redacted there; the credential itself is still hidden wherever it appears on its own.
• Redaction is best-effort output hygiene, not a security boundary.