I don't see why not, though a strict RAW approach suggests no.
The rules seem unclear, but if you want to go with an ultra-strict RAW approach then there is zero ambiguity in the multiattack description: if the Bandit Captain doesn't have a dagger, they can't complete the written sequence of attacks. Since there is no conditionality or separability in the text, all-or-nothing is the most RAW-conforming ruling.
However, that conclusion is at odds with other applications of multiple attacks in a turn. Aside from the use of and in the description, there isn't any reason to assume the dagger is important to the preceding scimitar strikes-- various situations can be argued (maybe the dagger is part of some practiced pattern of strikes and parries), but all rely on moving beyond the text. Consider the following four scenarios:
- A Bandit Captain only has a scimitar equipped, but mimes holding a
dagger in their off hand. The third attack (with the mimed dagger) wouldn't be very
effective, but would that matter for the attacks made with the weapon
the captain actually is holding?
- A Bandit Captain is in a tricky spot, and while not carrying any
daggers has fashioned a shiv out of materials at hand. The shiv is an
improvised weapon (and so is not properly a dagger), but is otherwise
dagger-like in size, weight, and use. This shiv is like a dagger in all game-mechanically-defined ways except its name and proficiency category. This would run afoul of the
text (it's not a dagger), despite creating a situation nearly
identical to what the text describes.
- A Bandit Captain is fully equipped with scimitar and at least one
dagger. Different attacks within a multiattack are separable, and so
can target different things within that one action. This Bandit
Captain, for some reason unimportant to this situation, chooses to
target a PC with both scimitar strikes and then empty space with the
dagger. This is allowed, but clearly neither a meaningful target nor
intent to hit and cause damage are required for the multiattack
action to occur. And so if an effective attack with the dagger is
not a requirement, should a situation where that factor alone is
forced (due to not having a dagger at all) be considered so different
as to disable the entire action?
- A PC Fighter of sufficient level has a multiattack class feature
enabling three attacks per turn. They can, however, opt to only
make one or two attacks should they choose to do so. While there are
obviously differences between a PC with class levels and an NPC
opponent without, is this weapon-restricted multiattack description
so different from the multiattacks available to PCs that the Bandit Captain does not have the flexibility to attack less than three
times?
Notes:
Ammunition is odd for NPCs (and PCs too, frankly). There isn't any reason to insist that Bandit Captains carry some specific number of daggers at all times. They might have zero, or one hundred. Unless a table is faithfully tracking ammunition amounts, there's not reason to assume that any dagger is a Bandit Captain's last dagger no matter how many are thrown. The situation in the question is interesting, but is an edge case that may not ever actually come up at the table.
Tactics for a Bandit Captain would probably change depending on how this is ruled. A Bandit Captain who throws their last dagger and then drops from three melee attacks to one is a captain that will probably not throw their last dagger. Possibly dealing 1d4 damage is a poor tradeoff for future attacks which would otherwise include that same chance of 1d4 damage and also an additional scimitar strike. Multiattack is one of Bandit Captains' major features, and losing it for such a weak attack is nearly always going to be dominated by other options.