You and Jin both expressed issues with your hardware. Probably others as well have experienced these issues.
Microswitch joysticks seem to generate a "signal bounce" which disrupts the TIA on select early 6-switch Atari units. Later hardware were not affected.
When the microswitch is actuated or released, the signal does not turn on or off instantly. When you view this signal on a scope or logic analyzer, there is not a clean transition from on to off and visa versa, but a lot of noise which lasts for a couple microseconds or so.
The TIA reads directly from the paddles, Fire buttons, and difficulty switches. The CPU reads from the joystick inputs. The Fire button connects directly to the TIA with little to no buffering, and the TIA momentarily outputs garbage on random scanlines correlating to the exact instant that the button is pressed or relessed.
What is not affected: Controllers with dome contacts (ie vintage cx-40 joysticks) or rubber/silicon carbon dot membranes (ie modern Atgames replacement joysticks), vintage or custom 7800 controllers with inline resistors for two button support. Genesis controllers. Later revisions of Atari hardware.
What is affected: Direct wired 2600 controllers with microswitch joysticks and buttons, vintage 3rd party sticks consisting of a bare metal shaft which contacts "leaf" actuators (ie tac-2), when paired with certain revisions of 6-switch Atari systems.
I have proposed a mod [untested] which may fix the issue on affected consoles by connecting a small value ceramic disc capacitor between pins 6 and 8 of the joystick port. I don't know what range of values might work but my educated guess is to start in the tens to hundreds of picofarads ballpark and increase the value until the problem resolves itself. Signal bounce should only last a couple of microseconds on a quality microswitch with snap action, so a smaller value would be preferable. Too large (ie greater than around .1uF) and it will take some time for the cap to recharge against the pullup resistor inside the console, resulting in the button input being sustained for a longer duration than the actual button press.
This is only a theory at this point, but I have a gut feeling adding a cap over the fire inputs inside the console may suppress any unwanted signal bounce. Other than that, I'm out of ideas. If someone wants to loan me their console (that exhibits the TIA glitches when using a joystick with microswitch fire button), I can try this idea to see if it resolves the issue. I've literally got piles of custom built jousticks to test with. For testing purposes, the capacitor will be added across the joystick Fire button. If it successfully suppresses the graphical artifacts, I will install it inside the console, for free, with owner's permission. The fix will be documented in a video for others to perform. I will even pay return shipping.