This fix avoids a WS_Alloc panic when pushing fetch processors
and should thus avoid panics by overflowing the backend workspace in vcl.

This overflow is logged as FetchError "Bo workspace overflowed".

Other panic points due to workspace_backend being set too low still exist.

Sizing estimate for workspace_backend:

	sizeof(struct busyobj)
	+ 3 * HTTP_estimate(cache_param->http_max_hdr)
	+ cache_param->vsl_buffer
	+ cache_param->http_resp_size
	+ fetch processor memory
	+ space required in VCL

Fixes #1739