Fix `{{..}}` templating -- standardize on `expr`

The current templating system appears to be the primary cause of confusion and bugs. These are caused by the following:

1. Template evaluation is unpredicable. By which I mean, you don't know when it'll evaluate and what the output will be. This can result in a workflow that passes just some of the time
2. We use several evaluation technologies: Govaluate, fasttemplate, Expr+Sprig.

Minor issues:

1. `{{..}}` does not play nicely with Helm templating.
2. `podSpecPatch` is arguably a work-around rather than a solution.

It would be great to remove this entires class of issue.

To fix this we should:

* Standardize on `expr` as the templating technology (easy).
* Make evaluation time predicable - at field access time (hard).

The issue names no files, tests, or entry points. Start by mapping the existing Govaluate, fasttemplate, and Expr+Sprig evaluation paths, then assess the scope of standardizing on expr and evaluating at field-access time; done means both goals are implemented with predictable templating behavior.

Written by the indexing model from the issue text.