`--strict-equality` too strict in iteratively visited code

[tyralla]

The second error report is good, but, in line with the decision made in #6735, the first one is not:

# flags: --allow-redefinition-new --local-partial-types --strict-equality
x = 1
while True:
 if x == str(): # error: Non-overlapping equality check (left operand type: "A", right operand type: "C")
 break
 x = str()
 if x == int(): # error: Non-overlapping equality check (left operand type: "str", right operand type: "int")
 break

This could be easily fixed by letting the IterationErrorWatcher also handle COMPARISON_OVERLAP errors (which would already be an improvement over the current handling of functions with constrained type variables of just turning off --strict-equality temporarily). However, IterationErrorWatcher would need some adjustments to not accidentally filter out errors with slightly different texts, like in this case:

# flags: --allow-redefinition-new --local-partial-types --strict-equality

class A: ...
class B: ...
class C: ...

x = A()
while True:
 if x == C(): # error: Non-overlapping equality check (left operand type: "A", right operand type: "C")
 # error: Non-overlapping equality check (left operand type: "A | B", right operand type: "C")
 break
 x = B()

Ideally, Mypy would only emit the last report (with A | B), which could eventually be implemented similarly to what is suggested in #19324 for reveal_type.

[gschaffner]

FYI: #20294 looks similar to this and is probably related, but I reported it separately since it seems to only occur in some specific conditions that aren't present in the reproducer above.