tradBoth are defensible, and the choice depends on what you want the journal to do.
IMPORTING EVERYTHING gives you something to measure immediately. It is the only way to have a usable sample on day one, and therefore to answer questions that need volume: do I exit too early, is my win rate stable, what do my losing streaks look like. Without history, those answers wait months.
STARTING FROM ZERO gives a cleaner read of what you do NOW. If you recently changed method, instrument or size, an average reaching back two years mixes two different traders and describes neither.
WHAT WE MOSTLY DO, which avoids having to choose: import everything, then DATE the change. Statistics are then filtered on the period that matters. You keep the depth for questions that need it, and you look at the recent period to judge your current method.
THREE THINGS TO KNOW BEFORE IMPORTING, which save doing the work twice.
DUPLICATES. Importing the same statement twice creates the same trades twice, and every statistic silently becomes wrong. Check the trade count after import against the statement.
DEMO. If your history mixes demo and live, separate them: demo trades inside a live account average inflate everything computed on it.
FEES. Depending on the export format, commissions and swaps are included or sit alongside. A history imported without fees gives a P&L that is systematically too good, and the gap grows with the number of trades.