Don't require leading | for Zed that's applied after inclusions

[philrz]

Repro is with zed commit 969f37d.

Imagine a user has some Zeek NDJSON data and they want to bucket addresses into subnets. They first try to process the data without the assistance of any shaper.

$ zq -z 'put classnet=network_of(.id.resp_h) | count() by classnet | sort classnet' weird.ndjson.gz 
put: a referenced field is missing
...

At this point they realize "oh shoot... this NDJSON has no rich typing!" To turn id into a nested record that includes the ip-typed fields like resp_h, they modify their command line to to apply the reference shaper for Zeek, but now receive a parse error:

$ zq -z -I ~/work/zed/zeek/shaper.zed 'put classnet=network_of(.id.resp_h) | count() by classnet | sort classnet' weird.ndjson.gz 
zq: parse error: error parsing Z at line 114, col 1:type port=uint16;

The problem is that the shaper ends with put .=unflatten(.) | put . = shape(schemas[_path]), so for our additional put to succeed, the user needs to prepend a | in their additional Zed like so:

$ zq -z -I ~/work/zed/zeek/shaper.zed '| put classnet=network_of(.id.resp_h) | count() by classnet | sort classnet' weird.ndjson.gz 
{classnet:"not an IPv4" (error),count:3 (uint64)} (=0)
{classnet:8.0.0.0/8,count:1 (uint64)} (=1)
{classnet:10.0.0.0/8,count:23889} (1)
...

It seems it would be helpful if the platform could figure out on its own when the | is necessary to stitch together the additional Zed top of the included Zed so the user doesn't have to make this adjustment.

[philrz]

@mccanne noted this should be easier after his hazmat PR merges.