If I need current mirrors, can I just use dual NPN packages? I know that if multiple BJTs are manufactured on one die, they can pretty much be guaranteed to have the same spread.
Should I just stay away from them and just go with those that are labelled as "current mirrors?"
It's just hard to find a current mirror with the right Ft.
Of course you can use a dual NPN for a current mirror. The question is only how accurate the mirror will be. The better matched the transistors are, the more accurate the current mirror.
Using two transistors in the same package is usually better than two discrete transistors, even of the same model. One reason for putting two transistors in a package is for better matching. In that case, the datasheet will tell you not only the absolute range of parameters like gain and B-E voltage, but the maximum variation between the two parts.
Unless you really need the current mirror output to have the lowest possible voltage drop, you can make the mirror more accurate by giving each transistor its own emitter resistor. Now the overall function of transistor current due to input voltage becomes more predictable and matched between the two. The input voltage is now from the base to the other end of the emitter resistor, not just from base to emitter as with a bare transistor. As the emitter resistor increases, the gain and B-E drops of the individual transistors become less relevant.
Here is what I'm talking about:
This is usually a good idea when making a current mirror from discrete parts. On a IC, the transistors are well matched and resistor are hard to make. In a discrete design, transistor matching is not nearly so good and resistors are cheap and available.
To size the resistors, you need to know what the maximum current will be and what maximum voltage drop you can tolerate. Let's say you want a mirror to work up to 10 mA and 2 V overall drop is acceptable. Let's say you want to keep 1 V across C-E of each transistor to keep it operating in the current sink range. That leaves 1 V across the resistors. By Ohms law, the resistor values are (1 V)/(10 mA) = 100 Ω. Such a current mirror will be quite accurate with any two transistors of the same model out of the bin.
Note that the two resistors need not have the same value. The Iout/Iin current gain is R1/R2 when the transistors are ideal. In the example above, you could have Iout be 0-10 mA as before, but controlled by 0-5 mA in with R1 200 Ω instead of 100 Ω
Generally speaking you should not depend on what is not in the datasheet.
Transistors in duals (and those side-by-side in the SMT tape) do tend to be matched much better than those selected from different batches or altogether different manufacturers. I have done relatively informal measurements of small lots. The typical matching is relatively good, but common sense tells you that eventually you will get a pair that came from different edges of the wafer, a different wafer, a different wafer lot, or even a wafer from a different facility.
There have been some attempts to sell relatively inexpensive pairs with guaranteed matching characteristics (VBE and hFE) but I am not sure that the market has proven sufficient to maintain production. Eg. BCM846S/BCM856S. They are not recommended for new designs at Infineon, and I see little to no stock from NXP.
The expensive matched pairs have also decreased in availability, and as you say they tend to have low fT figures, being aimed at precision low frequency applications.
In some situations you can use op-amps or emitter degeneration to reduce the requirements for matching, of course, but I assume you've already considered that. There are applications such as log/antilog amps where it's not much help.
Edit: There is a line of matched transistors from Diodes Inc. that you may find useful.
Built with adjacent die from a single wafer: DC Current Gain, hFE, VCE(sat), VBE(sat) are matched to a 2% maximum tolerance.
See Matched Pairs | Diodes Inc., and if that URL disappears, I've reproduced the list below. There seems to be plenty in stock (300k+ at the factory for one type), and they are very, very inexpensive compared to any kind of semi-automated or manual testing.
When you need a current mirror, the thing to look for is not just dual transistors (though any dual package will be better than two separate ones): you want what's called a matched pair. These are a bit less common, but for most purposes the still-readily-available BCM847 (NPN) and BCM857 (PNP) are handy (as a bonus, they are also available in the tiniest little 1 mm by 1 mm package, for space saving).
You can find Digikey's matched pairs here; note that some of these are five-pin packages with a common emitter terminal, so they can't be used for all purposes you might want a matched pair for. There are also a couple of (very expensive) matched quads, though the use cases for these are more specialized.
Matched pairs are usually fairly low-current transistors. You're not going to find any 30 A power parts in here. The highest they get are the BCM53 (NPN) and BCM56 (PNP), both rated to 1 A collector current, but the majority of them are in the range of 100 to 200 mA.
It is also possible to get matched pairs of MOSFETs, should you need that, but availability is worse (there appears to be only one supplier, and they're not as big a name as Nexperia or Diodes) and they're much more expensive (it's not as easy to make matched MOSFETs as it is BJTs).
