Designing a safe multimeter-like oscilloscope requires both channel-to-power and channel-to-channel isolation. It's certainly possible, but it can be awkward and difficult, especially if the oscilloscope has multiple channels and is powered from the AC mains. It's much easier to design an Earth-referenced oscilloscope, as it allows you to ignore the engineering problems involved in both kinds of isolation.
First, oscilloscopes must be shielded to protect themselves from high-frequency interference. The textbook way is creating a continuous Faraday cage by enclosing all circuits inside a metal chassis (or at least with metal covers above PCBs), all coax input connectors must also be screwed onto the chassis with a 360-degree contact. Without a connection to Protective Earth, the chassis and all metal parts create an obvious electric shock hazard, so it would be necessary to use plastic covers above all metal parts, and use safety-certified, insulated versions of coax connectors without exposed metal, which is non-standard.
But the problem goes further than that. By sharing a common shield, all signal grounds are automatically joined together via the chassis. If the ground of one channel is energized, all other channels are energized too. The hazardous voltage from a single high-voltage channel can propagate to other channels, an innocuous-looking low-voltage coax cable (with exposed metal) on another channel may become dangerous unexpectedly. To avoid this safety problem, the oscilloscope can be made to accept non-standard cables only, but this makes the oscilloscope rather useless, and still doesn't allow one to make two independent measurements with different Common references. So the alternative is to isolate the input channels.
Then, isolating input channels creates its own challenges. To my best knowledge, off-the-shelf isolation amplifiers are designed for low-frequency analog signals, not RF signals, so the solution requires a custom front-end. One may say that differential probes are routinely used for these kinds of measurements, but they're internally just differential amplifiers with a very high common-mode impedance (the equivalent circuit is two megaohm voltage dividers followed by an ideal differential amplifier), and are not certified for isolating hazardous voltages, in the same sense that the AC mains is still unsafe to touch behind a voltmeter, even if the voltmeter has 10 megaohm series impedance. Shielding is also problematic, since the oscilloscope can no longer be treated as a simple continuous Faraday cage. In particular, coax cables are problematic, as they use same conductor for signal ground and RF shield, so they really want you to join the signal ground and shield together by design (Due to skin effect, the single conductor is effectively two conductors, the return current flows on the inside, the noise current flows on the outside. This is why joining coax shields and signal grounds is the standard in RF systems (to the dismay of audio engineers).
I imagine that one possible solution to both problems is to design each input channel as its own island, with its own circuit ground, shield and own front-end circuit. This pre-conditioned signal can possibly be sent to the next stage using RF signal transformers (but are there any RF transformers with safety-rating for mains voltage?). Another solution is using a dedicated ADC for each channel, and isolating the digital interface. Some expensive high-voltage probes even use optical isolation, which theoretically can be used here too. Furthermore, an even more radical approach would be stop using coaxial cables altogether, and instead using unshielded twisted pairs which possibly have acceptable EMI rejection for signaling and cabling without forcing you into the decision of sharing signal ground and shield. Other ideas like shielded twisted pair or triaxial cables are not out of question if you're willing to break with tradition, but these designs are either incompatible with existing equipment or impractical for probes.
Finally, power isolation is also problematic. As long as the oscilloscope is powered from the AC mains, the oscilloscope's input channel can never truly be isolated from the Protective Earth of the AC mains due to parasitic capacitance between input channel and power, furthermore Isolation transformers (in AC/DC power supplies, or in DC/DC converters) are far from perfect due to parasitic capacitance. Thus, even if every signal and power goes through isolators, Common-Mode Rejection Ratio is still an important engineering consideration since it quickly degrades at high frequencies due to parasitic coupling between adjacent channels, and between input channels to the Protective Earth. To make matter worse, there are conflicting EMI requirements - when two conductors have different AC voltages, a dipole antenna is formed, radiating electromagnetic interference from the switched-made power supply. Thus, this often requires us to intentionally increase the coupling capacitance across the isolation barrier via Class-Y capacitors.
None of these problems are unsolvable, it's just an engineering question. But nevertheless, Earth-referenced oscilloscopes is a standard and simpler design with more than half a century of history, and the problem of differential measurements are satisfied by differential probes anyway. This makes fully-isolated oscilloscopes a specialty at premium prices.
For multimeters, these problems are easier to solve. Most multimeters only have a single input channel, thus problems of channel isolation is bypassed. The use of batteries completely bypassed the problems of AC mains isolation. Many isolated oscilloscopes are battery-powered for the same reason. Furthermore, multimeters use banana jacks to handle low-frequency AC signals, so the probes are rarely shielded. This bypasses the problem of sharing shield and signal ground for coax cables. In fact, dedicated "shield" or "guard" connections only exist in special electrometers or metrology-grade multimeters, and don't exist in mid-range benchtop multimeters - if you're doing a sensitive experiment, you have to shield the meter yourself...
