Well, the first thing to note is that many types of music analysis attempt to create "reductions" of music that privilege certain notes over others. Whether you find those analyses compelling is to some extent up to your own judgment and ear, as well as whether you find them useful for understanding the music.
We do know from historical treatises in the baroque and classical period that composers learned about the art of "diminution" where they'd elaborate a particular simple melody (often moving stepwise) with other notes of shorter duration. Effectively, they learned to start with a "skeleton" of a few basic notes and build up ornaments and gestures around them. But there's a lot of controversy about exactly how to "reduce" a particular collection of notes to that underlying "structure." Since music is an art, not a science, there's no consistent set of algorithms you could give to reduce all melodies this way in a consistent and also musical fashion. (Believe me, many theorists have tried.)
The best we can do is try to look for lines that seem to tell us something about the structure of the overall melody at hand. In cases like Kennan analyzes (both here and in his other examples), he's also dealing with compound melody, that is, a melody that appears to be imitating two or more individual voices while leaping between them.
That's what this melody is an example of. From the beginning, it seems to split into two voices. The upper voice keeps doing turns: C-B-C...C-B-C-D...C-B-C-D, which are based on that opening neighbor tone motive. Meanwhile, the leaps down consistently move to another voice that appears to have its own melodic coherence: G-Ab...G...F-G-Ab-G-F-Eb. It too has a bit of a rise and fall, articulating the G-Ab-G important to establishing the minor mode in baroque melodies, just as the upper line establishes the tonic and leading tone by repeatedly moving C-B-C.
(Note: I somewhat disagree with Kennan's consistency when he states that there is only one step progression here, as in later examples he allows neighbor tone motion to be classified as a separate voice creating a step progression. That's what I'd argue is happening here in the upper voice.)
Composers very commonly thought of compound melodies in this fashion, with each "voice" moving according to the normal rules of voice-leading and counterpoint.
Kennan wants to go one step further with that bottom line in reducing it to its general direction, which is to gradually fall downward. Hence his Ab-G-F-Eb line. It's important to understand that his "gaps" between notes mean somewhat different things here: He doesn't include the C-B-C-D notes because they are articulating a different "voice," but he skips the later G-Ab-G-F at the end of the second bar because it's more of an elaboration ("diminution") of the overall line trending downward.
Anyhow, whether you find this analysis compelling or useful depends on your aims. In glancing over Kennan's other examples, he seems to privilege this notion of compound melody in his search for step progressions. Other analysts might be more or less lenient in the way they find these "lines." What is definitely true, though, is that actual composers would have considered the counterpoint created by compound melody. So whether you actually "hear" these lines or not, they often do have something to do with the way the music was composed.
In sum, a few reasons why Kennan's line of Ab-G-F-Eb seems compelling to me as an underlying structural element of this melody:
- The notes he chooses all occur on metric strong beats.
- The notes he chooses are all in the equivalent places metrically within the beat structure.
- The notes he chooses are accented several times by leaps to or away from them, usually an act reserved in baroque counterpoint for notes that were important to the melodic/harmonic structure. (Composers of this time very rarely would leap to/from random dissonant or unimportant notes, except when part of an idiomatic gesture.)
- The line created by these notes follows one element of the overall trajectory of the melody, which feels like it falls down by its end.
Reasons why OP's postulated stepwise line (A flat, B, C, D, C, B, A flat, G, F, E) doesn't seem as compelling to me:
- The line requires augmented seconds in two places, which were rarely used by composers at this time in melodies. When they do occur melodically, they are frequently part of a "compound melody" gesture articulating separate voices.
- The line arbitrarily chooses notes from any metric position.
- Even aside from the augmented second, the line doesn't take into account contrapuntal elements of compound melody suggested by the leaping, nor does it adequately relate to the use of the standard tendency tones in minor keys (8-7-8 and 5-b6-5) that are present in this melody and are an important part of baroque counterpoint.
I agree that when looking at analyses of this sort, sometimes choices can seem arbitrary. And sometimes, frankly, I would agree with you that they are. However, one also needs to understand the type of analyst one is dealing with and what the assumptions are. Kennan seems to be using his "stepwise" skeletons to show the formation of different voice parts in compound melodies that may be present. And most of his decisions seem pretty consistent to me in that regard, but sometimes he seems more intent in finding a scalar descent or ascent, so his choices are a bit more arbitrary. While one can fault his consistency, it's clear that the melodies he's looking at tend to have an overall direction to them, and his analyses show the directionality of the underlying contrapuntal lines.