"Well as we are measuring a trend, it doesn't really matter how the error bars are distributed as long as we don't think that there is a systematic change in distribution of error over time."
That's only true if you don't care about slope, acceleration or deceleration, since they depend on correct magnitude and linearity.
In addition to that, instruments can have offset scales, drift, random noise or hysteresis. All but the last two can be directional, and all six can be caused by manufacture and design.
Drift can can fabricate a trend entirely, as can changing from one class of instrument to another with a different error profile.
"If you think there has been [a change in distribution of error over time], you need to justify why"
Science requires traceability and reproducibility, which you can't claim of a dataset that depends on undocumented behavior of unnamed instruments. I'm only obligated to rebut scientific conclusions.
(There are also plausible sources of bias in manufacturing, such as compression of the capillary of a liquid-in-glass thermometer during the glass-blowing step if the glass heats up more than the fixture that holds it.)
"...and how come that change has been gradational rather than stepped"
Non-linearity is as likely to be smooth as stepped, and magnification is evenly distributed over an instrument's range by definition. (The same physical defect may cause other types of error, but they're accounted for separately.)
Switching to a different type of instrument would only cause a sharp change in the global trend if all of the international agencies supplying data made similar changes simultaneously.
"...and why the change has the same sign for different measuring technologies."
As far as I'm aware, it would only take one type of instrument to skew the GST, since lower readings at sites with other types would be attributed to local variability. It's possible that the SST is more resistant to that.
I can't comment on satellite measurements except to say that they would need to be prepared much more thoughtfully than the surface temperature record.
"For extra credit explain why temperature surrogates (tree rings,, stalactites etc) are congruent with the instrumental record."
I've only read Anderson et al. 2013, which doesn't include tree rings:
"To avoid potential loss of low-frequency variations, we excluded tree ring series."
Most of the site records relied on O-16/O-18 ratios:
"Sixty percent of the 170 records are oxygen isotopes preserved in ice or carbonate"
Precipitation has been increasing (as per climatologists' own literature), and snowfall has been underreported due to problems with gauge design (ditto). Precipitation decreases salinity and increases the O-16 fraction, both of which would be interpreted as increasing temperature in Anderson's index.