chain functions broken due to oa prefix in new batch-machine

[andrewharvey]

https://github.com/openaddresses/openaddresses/blob/master/ATTRIBUTE_FUNCTIONS.md#chain documents use of chain functions with a _wip variable, the example given is:

"street": {
    "function": "chain",
    "variable": "street_wip",
    "functions": [
        {
            "function": "postfixed_street",
            "field": "Prop_Addr"
        },
        {
            "function": "remove_postfix",
            "field": "street_wip",
            "field_to_remove": "Prop_Addr_Unit"
        }
    ]
}

This used to work fine with https://github.com/openaddresses/machine but once that we deprecated in favour of https://github.com/openaddresses/batch-machine a breaking change was introduced which meant this example would now fail as evidenced by the errors in openaddresses/openaddresses#5496

After a few months of not understanding the issue, I took another look comparing the old conform https://github.com/openaddresses/machine/blob/05db17d8492b3d8f4064f0f5b0ca9c68041c535a/openaddr/conform.py with the new one https://github.com/openaddresses/batch-machine/blob/8bb6ecfb20beec56913ae7ae267cf5489c9c35ad/openaddr/conform.py this revealed an issue at

batch-machine/openaddr/conform.py

Lines 1200 to 1218 in 8bb6ecf

	def row_fxn_chain(sc, row, key, fxn):
	functions = fxn["functions"]
	var = fxn.get("variable")
	original_key = key
	if var and var.upper().lstrip('OA:') not in sc.SCHEMA and var not in row:
	row['oa:' + var] = u''
	key = var
	for func in functions:
	row = row_function(sc, row, key, func)
	if row.get('oa:' + key):
	row[key] = row['oa:' + key]
	row['oa:{}'.format(original_key.lower())] = row['oa:{}'.format(key)]
	return row

The code here now creates these intermediate variables with the oa: prefix, once I added this prefix to the _wip variable fields at https://github.com/openaddresses/openaddresses/pull/5828/files#diff-8441ea2e82820b7decef018f2d606c8a543ad86411a2c49aedab254a0f72f848 it worked.

So the question for you @ingalls is, is this either

a) a big in batch-machine that we can fix, or
b) a change in how these conform files are processed and we should update the documentation and existing sources which broke

I'm happy to help out, but I need to first understand some of the background and motivation of this change.