I can recommend using Polar (https://github.com/jamaljsr/polar) to roll out a local LN. It runs real daemons such as Bitcoin Core and LND and they connect to each other using real sockets. You can insta-mine btc and give it to all the nodes, open channels, create and pay invoices etc., the usual stuff. Everything is instant, free and as close to the real network as possible.

No real way to answer that, just open some channels and see how it scales. Liquidity only limits payments in one direction, i.e. you can't send more than 1M sats through a 1M channel. But if you send and receive back, you can do that indefinitely long. So what actually matters is the balance between selling and buying goods&services between those businesses.

I think a good rule of thumb would be opening a channel with at least 3 times greater capacity than your monthly turnaround so you'd have some margin. It's possible to buy inbound with push, i.e. you buy a 3M sat channel and your balance will be 1.5M (of course, you pay that 1.5M on chain + the service fee for 1.5M inbound). Now you can both spend and receive which is ideal for a business because if all businesses open channels with no inbound they can't pay because no one can receive, and if they all buy inbound they also can't pay because no one can send, and if they somehow agree that some of them open channels by themselves and some buy inbound it becomes overcomplicated and messy.

Another cool feature is that LN doesn't need to be global or even public. Businesses, banks and communities (such as villages) can run their own segments that don't interact with the main LN, fully privately. They don't even need internet for that except for opening the initial channels and monitoring the chain for force closes. Of course, they can also run their own Bitcoin network (and ditch the internet completely) but it wouldn't have any real value because it wouldn't be a global consensus money.