He must decide who(m) to be.
We can tell that “who(m)” in this sentence isn't the subject of "to be" because the meaning of the sentence requires the subject to be "he" (or an empty element with "he" as the antecedent). It means the same thing as "He must decide who he will be" and not the same thing as "He must decide who will be."
I think this sentence has the same structure as a sentence like
He must decide [to be an adult].
In other words, the pronoun “who(m)” is the complement of the verb "to be".
This corresponds perfectly with the meaning of the sentence.
The position of the pronoun “who(m)” in front of “to be” is regular due to the process of fronting. If we mark the original position of the pronoun with a “trace” t1, we get the following representation of the structure:
He must decide [who(m)1 to be t1].
Now, as mentioned in the question, both nominative and accusative pronouns sound odd after “decide to be” for semantic reasons.
However, if we compare this to other sentences containing "to be" with similar structure but different semantics, we find that the nominative seems to have been historically prescribed in this grammatical context. Correct English, Volumes 20-21, edited by Josephine Turck Baker, from 1919, says that the following sentences are "correct":
- It was supposed to be I who made the error.
- It was thought to be he to whom the speaker referred.
- I should like to be he/she.
I think this addresses Anonym’s point about “I want him to be here”: quite simply, Baker treats “to be” differently depending on whether it’s explicitly preceded by an accusative pronoun. And in a sentence like "He must decide to be an adult", there is no accusative pronoun, so Baker (and presumably other prescriptivists like her) would consider it incorrect to make the complement of "to be" accusative in this context.
As Rathony said in a comment, “The verb decide takes who to be as an object, not who itself.”
The fact that “decide” works this way is simply a fact about this particular verb in modern English.
Now, all of this rests on my original analysis of the structure of the sentence, which might be wrong (I'm not a syntax expert), although BillJ's answer seems to assume more or less the same structure I've presented here.
"Who" certainly sounds OK here, so even if I'm not 100% certain "whom" would be wrong, I'm confident enough to recommend "who" over "whom" here. As medica said in a comment,
Since whom is often misused as an overcorrection, the current
recommendation is, When in doubt, use who. That way, if you make a
mistake (either way), it will seem less pretentious.