Yes, this is the basic idea. Tokens from the lexer, AST nodes, basic blocks, control and data-flow graph nodes + edges can all be entities and/or components of entities, depending on how you wish to structure your data. Since a 32-bit entity ID provides a sufficiently large "address space", you get the added benefit of only needing to store 4-byte handles instead of 8-byte pointers.

EnTT has an eventing system as well. These can be triggered in various ways (entity created, entity deleted, component assigned/removed, etc). I've used eventing to control fine-grained compilation pipelines. For example a recent scheme compiler I wrote used events to simplify inlining (both define-macro/define-syntax and straight procedure inlining) and quote/quasi-quote/unquote/unquote-splicing translation.

If you spend a lot of time writing compilers, I find an ECS like EnTT can give you extra flexibility during the design/prototype phase. Try it out on a simple prototype and see if you like it.