The short answer
Do not start by cleaning records. Start by stopping the bleeding at the point of entry, because a cleanup on a leaky system just gets dirty again in six weeks. Pick the five fields the business genuinely runs on (account name, stage, amount, close date, owner), make them required or picklists, and delete or hide everything else so reps stop guessing. Then write a one page definition of what every pipeline stage means and get the sales leader to sign off in public. Only after that do the backfill: dedupe, merge, and fill the five fields on open pipeline first, closed history last or never.