POC: Expression builder class.

[lukeelmers]

[skip-ci]

The recent changes in #33773 got me thinking that until each saved vis object is stored as a true expression, our build_pipeline utility for converting a stored vis into an expression string is going to become increasingly complex and error prone, especially as we continue refactoring each of the functions.

I thought that perhaps we could reduce our likelihood for errors and eliminate some duplication by having a chainable helper class that could be used to help build the expression string. This a quick POC I put together; if it feels like we want to move in this direction, I will update this PR with tests & convert everything over.

const exp = new ExpressionBuilder();

// add a new function to the expression (functions are stored in the order they are added)
exp.addFn('kibana');

// add a new function with some arguments
exp.addFn('tagcloud')
  .addArg('visConfig', { foo: 'bar' }); // will handle stringifying & escaping objects

exp.addFn('render')
  .addArg('baz', '"hello"'); // "hello" gets escaped

// whoops, forgot to add dimensions to the tagcloud function earlier
exp.selectFn('tagcloud')
  .addArg('metric', '{ vis_dimension 1 }'); // works when given a subexpression as a string

// print an individual function
exp.printFn('tagcloud');
"tagcloud visConfig='{ foo: \'bar\' }' metric={ vis_dimension 1 }"

// print the whole expression string
exp.print();
"kibana | tagcloud visConfig='{ foo: \'bar\' }' metric={ vis_dimension 1 } | render baz='\"hello\"'"

Basically the class takes care of the following:

1. Storing provided functions in order (FIFO)
2. Adding spaces between args & pipes between functions
3. Escaping strings properly
4. Stringifying JSON when provided as an arg
5. Skipping any undefined args - currently we have a ton of if (foo) pipeline += foo type of checks in the code that this would eliminate the need for
6. removing or updating functions
7. removing or updating args on a function

Maybe a utility like this would even have applications in other places, like unit tests? idk

cc @ppisljar

[lukeelmers]

Closing as we agreed long ago that we'd prefer a utility to manipulate the AST directly (rather than manipulating strings, which don't have type safety). This work is being tracked in #56748