Testing patterns on a one minute historical basis is fine. Live not so much. And no way for entries/exits. Need ticks for that, otherwise you play the old assume game. "I bought at the low of the bar and sold at the top of the bar" Right!!!
Yes. That's why I tend toward Polygon. I need to find out whether it would be possible to access their API with multicharts. Polygon and IQfeed both offer > 10 years of minute data but IQfeed only offers 180 days of tick data. Polygon seems to offer 10-15 years of 1 second data which would totally suffice for my purposes. As you point out it is going to be very difficult to estimate whether the backtesting results simulate the reality adequately or not... My idea/hope is that a strategy with many small trades might/should leed to more confidence in the strategy (law of large numbers).
I do all the above. Depends on the need. Can do many years of tic data on all exchange traded stocks with an inexpensive used Dell rackmount server running Linux. You obviously want to use the least amount of data required. If you only need 1-min or 10-min bars then use that. If you need to analyze every trade and NBBO change then do that. Use a relational database when it makes sense to do so. This just isn't a thing to worry about. Start writing the software and go do it. If you can't get enough CPU cycles to run your tests then you are doing something wrong. A used Dell server with 24 cores is inexpensive and plenty powerful.