fix: placeholder arrays in record __repr__

[pfackeldey]

Follow-up of #3341
This fixes the __repr__ for Record arrays that contain PlaceholderArrays:

import awkward as ak
import numpy as np
from awkward._nplikes.numpy import Numpy
from awkward._nplikes.placeholder import PlaceholderArray

numpy = Numpy.instance()

form = {
    "class": "RecordArray",
    "fields": [
        "foo"
    ],
    "contents": [
        {
            "class": "NumpyArray",
            "primitive": "int64",
            "form_key": "node1"
        }
    ],
    "form_key": "node0"
}

# without this PR
ak.from_buffers(form, 3, {"node1-data": PlaceholderArray(numpy, (3,), np.int64)}, highlevel=True)
# ... TypeError: PlaceholderArray supports only trivial slices, not int

# with this PR
ak.from_buffers(form, 3, {"node1-data": PlaceholderArray(numpy, (3,), np.int64)}, highlevel=True)
# <Array [{foo: ??}, {foo: ??}, {foo: ??}] type='3 * {foo: int64}'>

Another cool thing: It can show mixtures of e.g. PlaceholderArrays and numpy (or others) now too:

form = {
    "class": "RecordArray",
    "fields": [
        "foo",
        "bar"
    ],
    "contents": [
        {
            "class": "NumpyArray",
            "primitive": "int64",
            "form_key": "node1"
        },
        {
            "class": "NumpyArray",
            "primitive": "int64",
            "form_key": "node2"
        }
    ],
    "form_key": "node0"
}

containers = {
    "node1-data": PlaceholderArray(numpy, (3,), np.int64),
    "node2-data": np.array([4,5,6], np.int64),
}
ak.from_buffers(form, 3, containers, highlevel=True)
# <Array [{foo: ??, bar: 4}, {...}, {...}] type='3 * {foo: int64, bar: int64}'>

[jpivarski]

Wait, so this was already in

awkward/src/awkward/prettyprint.py

Lines 79 to 81 in 35d8391

	def get_at(data: Content, index: int):
	if data._layout._is_getitem_at_placeholder():
	return PlaceholderValue()

and I just overlooked the get_field case in #3019? That's just my mistake, then. The get_at and get_field are supposed to be parallel to each other; one for rows and the other for columns.

It is definitely the right fix.

[pfackeldey]

Yes, that's my understanding aswell. 👍