This commit fixes the following problems, whic occur if a ViewBox with locked aspect ratio is resized:

1. The visible range changes suddenly at some points.
2. The set limits are not taken into account.

Note, this PR fixes also Issue #2415. (Thanks @aksy2512 for #2540, I took some ideas from your PR.)

You can reproduce this with the mwe below:
Run the script and then slowly decrease the window width.
You'll observe, that the minimum visible y value will go beyond -5, which is set as yMin limit.
This is Issue 2.
If you continue to decrease the width, shortly after the y-axis labels 2 and -6 become visible, the y-range suddenly changes to around -4 ... 0, which is the above mentioned problem 1.

To improve traceability, I try to provide some info on why I changed what in the following:
This is, because my fix doesn't change one simple thing. Also I had some difficulties to understand the updateViewRange() function and still doesn't understand everything, especially the part within if aspect is not False ....

• Adjusting viewRange to limits in updateViewRange() is necessary to fix the limits issue (2).
• That adjusting targetRange, when changing viewRange is necessary shows me the mwe from ViewBox.setLimits() doesn't trigger update if view range not within new limits #2415:
  There, it all works fine if you increase the width of the LinearRegionItem, i.e. the right plot gets updated if necessary.
  But if you then decrease the width of the LinearRegionItem again, the right plots x-range also decreases! This is prevented by updating the targetRange here.
  • But, we must not reset all targetRange parts (i.e. simply call self._resetTarget()) since this causes the following problem:
    If you start decreasing the width of the LinearRegionItem again, you observe that the right plots x-range just slightly decreases before stopping!
    You might adjust the mwe and print the x-range on a change, i.e. add p2.getViewBox().sigXRangeChanged.connect(lambda vb, range: print(range)) to see this effect.
    This doesn't happen if we only adjust the targetRange parts whose corresponding viewRange parts we changed.
• Regarding self._resetTarget(force=True) in resizeEvent(..), this fixes above described problem 1.
• I want to keep the influence of my fix as small as possible, therefore I added the argument to self._resetTarget().
  It would be nice if we could get rid of this if self.state['aspectLocked'] is False completely.
  Maybe someone knows a case, where this is needed?!

Minimum-working-example:

import numpy as np
import pyqtgraph as pg

app = pg.mkQApp()
plot = pg.PlotWidget()
plot.show()

N = 10
data = np.eye(N) + np.flip(np.eye(N), axis=1)
for (n,m) in [(n,m) for n in (0,N-1) for m in (0,N-1)]:
    data[n,m] += 1.0
plot.addItem(pg.ImageItem(data, rect=[-N/2, -N/2, N, N]))
plot.showGrid(x=True, y=True)

plot.setLimits(xMin=-N/2,xMax=+N/2, yMin=-N/2, yMax=+N/2)

plot.setAspectLocked(lock=True, ratio=1.0)

plot.resize(400,200)
plot.setXRange(-5,+5)
plot.setYRange(-5,0)


if __name__ == '__main__':
    pg.exec()