Fix redaction regex reuse: avoid mutating patterns and keep allowlist working across calls

[SynthLuvr]

This PR fixes a subtle statefulness bug in the redactor caused by cloning regexes and forcibly adding the global (g) flag.

What changed

• Removed cloneRegex, which was creating new RegExp instances with g appended.
• Updated replaceAllMatchesWithContext and replaceBip39MnemonicMatchesWithContext to use the provided pattern directly in String.prototype.replace.
• Removed explicit "g" flags from the internal HEX / BASE64 / BASE64URL / BASE58 regexes and adjusted the mnemonic regex from "gi" to "i" to avoid unnecessary global state.

Why

Cloning and forcing g can make matching behavior unexpectedly stateful (via lastIndex) and can also change how callers’ patterns behave. In practice, this could cause allowlist rules to stop matching reliably after repeated invocations, depending on how the regex is reused.

Tests

• Added a regression test (allows whitelist each time) that calls the redactor repeatedly with a hex allow rule and asserts the allowlist continues to apply on every run.

Impact

• Redaction behavior is now deterministic across repeated calls.
• Allowlist rules remain effective even when the redactor is invoked many times in a row.