What Problem This Solves

Fixes an issue where Telegram bot credentials could remain visible in large log or tool-error strings when the credential crossed the 16,384-character bounded-redaction split.

Why This Change Was Made

The shared redactor now treats the two built-in Telegram credential patterns as full-context patterns. Those two linear regexes run against the complete string, while every other default pattern keeps the existing bounded-replacement behavior.

A focused regression covers both the Bot API URL form and the bare credential form with synthetic inputs larger than 32 KiB.

User Impact

Telegram operators no longer risk exposing a bot credential solely because it straddled an internal redaction chunk boundary. Normal redaction output and the performance behavior of unrelated credential patterns are unchanged.

Evidence

• Current main reproduction at 5aea34ad8c210ddba5a678d64c559da8414baea7:
  • 49 KiB synthetic Bot API form: raw credential remained unchanged.
  • 49 KiB synthetic bare form: raw credential remained unchanged.
• Patched direct probe:
  • Bot API form: leaked=false, output changed.
  • Bare form: leaked=false, output changed.
• Blacksmith Testbox tbx_01kx6h7my69dczc9yrfej5gabs:
  • corepack pnpm test:serial src/logging/redact.test.ts: 118/118 passed.
  • corepack pnpm check:changed: both tsgo lanes, core lint, and selected guards passed.
• git diff --check passed.
• Fresh structured autoreview: no accepted/actionable findings; correctness 0.86.