[ML] Improve bundles sizes related to shared code ml/transform

[walterra]

We are sharing code from the ML plugin (e.g. custom data grid components etc.) with the transforms plugin. Since the shared modules are exposed via e.g. plugins/ml/public this could affect bundle sizes, for example the ML page load bundle size might be bigger because of this even if the code isn't required for the ML plugin to load.

An example of this is the increased bundle size caused be this PR: #73561 — Because of extending the data grid component which is also shared with the transforms plugin, the ML base page load bundle grew in size.

Another possible optimization might be possible by explicitly exposing only types using import type { ... }.

[elasticmachine]

Pinging @elastic/ml-ui (:ml)