Is there a distinction between doubting the proposition that a sentence expresses, and doubting what proposition that sentence expresses? It seems to me that there is, but that is a distinction that threatens to be collapsing under your defence of KILPOD.

...There might be cases where it really is not clear even to experts whether the x should be classed as an ant or a termite, and there one might think that it is actually the meaning of the terms that is is being brought into question as well, but that is a very specific kind of case.

I think you are right that my defence of KILPOD does threaten to collapse that distinction, and that threat is present in Wittgenstein's own account in OC. It think there is a distinction to be made, but it is more a question of emphasis than anything clear cut. Donald Davidson's theory of meaning, for example, defines the meaning of a sentence in terms of its truth-conditions. To understand the meaning of "x is an ant" is to know when, and if, "x is an ant" is true. So, conversely, Davidson says: "to give truth conditions is a way of giving the meaning of a sentence" (Inquiries into Truth & Interpretation, pg. 24). Clearly, then (in so far as this is a plausible theory of meaning, which I think it is), the question of whether or not "x is an ant" is true, is not really separable from the question of what "ant" means - or, to be properly Davidsonian, what "x is an ant" means. One may look at x and say, "That is definitely a termite rather than an ant", but only because one already knows how to distinguish the meaning of "termite" from the meaning of "ant", in terms of the conditions which make "x is a termite" true. The only difference in the special case you mention is that the expert is dealing with a borderline case; the consequent difficulty in classification is both empirical and semantic, and might lead to a revision of the criteria for classifying ants and termites.

Borderline cases in the use of "I know that..." are of particular interest to W, with particular reference to what he dubs "Moorean propositions". Moorean propositions occupy an interestingly ambiguous area between the empirical and the non-empirical. An example would be "I have a head". Normally this would be treated as a non-empirical proposition, and would be senseless to doubt (in fact in normal conversation it would be a senseless thing to say). Normally, it would be senseless to say "I doubt I have a head", and consequently (according to W) it would be equally senseless to say "I know I have a head". But even here, one can imagine a special case in which "I have a head" is treated as an empirical proposition (i.e. by questioning it): a small child says "Do I have a head, daddy?", to which the reply is "Of course, everyone's got a head". As Andy Hamilton says: "A Wittgensteinian response to this...is that, here, the child is learning the meaning of "head", rather than enquiring whether she has one; indeed that there is no clear distinction here between learning facts and learning the meanings of words." (Hamilton 2014, pg. 203). So, strictly speaking, the logical possibility of doubt exists here, but that doesn't mean doubt is normal in this case. It only takes a special case for a logical possibility to exist. But Cartesian doubt goes much further than the Moorean propositions and arrives at a level of certainty that cannot logically be doubted. The Cogito takes us to the point where I cannot doubt my own existence (as a thinking thing) because the very act of doubting guarantees my existence (as a thinking thing). But W would say it is, for that very reason, nonsensical to say "I doubt I exist" and consequently equally nonsensical to say "I know I exist". The Cogito does not furnish us with knowledge or any foundation for knowledge. Cartesian "certainty" and its attendant scepticism about everything that doesn't have the same indubitable certainty, is the real target of KILPOD.