What not to do
What you are asking us, without knowing, is related to how to price a variance swap. Well, under a general diffusion process, variance swaps can be priced by forming a suitably weighted portfolio of options over a continuum of strike prices with the entire portfolio maturing on a given date. The intuition is that your exposure to volatility changes when the the spot price of the underlying changes for one option: in financial parlance, your vega is a function of the spot price. But for a pure volatility exposure, you'd like to get rid of that dependance.
The unfortunate thing is that if you move toward a model that admits conditional nonnormality in returns (in continuous time, a jump-diffusion model would do just that), you're demonstrably incapable of pricing variance swaps: you don't have a strategy that allows you to build pure exposure to volatility because quadratic variation is going to be polluted by higher moments (see Martin (2017) for details). I mention this obvious problem in case someone
What to do
On the other hand, there is something you can do which is valid, even under the general context of jump-diffusions. Variance swaps focus on the observed quadratic variation in the growth rate of log prices, so they're always polluted by higher order term. Martin introduced the idea of simple variance swaps (their payoff depend on squared price changes, weighted by squared futures prices) to build a new index. As it happens, just like the VIX is built by discretizing the integrals used in the pricing of variance swaps, his index is also built from a portfolio of European options on the S\&P500...
All you have to do, if you want to "go long volatility" is to look up Martin (2017), find the integral defining his index (the SVIX) and discretize it. You have a portfolio of options, just not weighed the same way as in the VIX. To determine how many options you need in practice, pick a few jump-diffusion models, run simulations and see how many options you need to get precise results. That method absolutely will give you exactly what you need to know to be long vega in as general a context as can be -- you know, outside stable processes where what you're asking wouldn't make any sense.