is there a flag like

--reset-index-on-initialization=false

Typesense is an in-memory search engine so it does need to reload index on restart.

During restarts , it reindexes which consumes lot of resources , causing failure in startups. DB size is ~3gb , 1.5M Json records spread across two collection. Ram is 4gb droplet VM How to make sure that without using too much resources it is able to restart

You have to use HA for production setup. And do rolling restarts.

but we have manual setup, for rolling restart to happen we need cluster setup right ?

Yes you need cluster