Because their deployment model is 'git push' your source code, they need to actively support a language in order to use Heroku at all. Every extra language they support directly translates to a larger market size. New, niche languages are disproportionately important to Heroku's target market, because they're frequently used by startups and hobby projects and Heroku's best use-cases are often for startups and hobby projects.

It's also not that much technical effort to support an additional JVM language: they already support runtime deployments on the JVM, so they just need to learn & automate a new build toolchain, and then the deployment artifacts are basically the same as for the JVM. I'd bet this portion of their stack is isolated from the runtime & auto-scaling parts, so they can assign a few engineers to it, build out the tooling, and then have it just feed into the existing JVM stack.

High profitability + low cost = it makes a lot of business sense to do it.