There's also shot (current) noise...
A 1 k-ohm resistor produces about 4 nV/rtHz ("every" engineer remembers this) or about 0.6 uV in a 20 kHz bandwidth. That does not seem like a lot but is about -125 dB relative to a 1 Vrms signal so resistor (Johnson) noise can be enough to help set the noise floor of some of the components Amir has measured. Note a MM phono cartridge input is typically 47 k-ohms so that can present a fair amount of noise. Records are noisy enough on their own so it is not usually a big deal, but using the wrong op-amp can exacerbate the noise. Low-noise op-amps using bipolar transistors have higher current noise due to their input (bias) current requirements and are not usually suitable for high-impedance inputs like phono stages; JFET-input op-amps are typically used. MOSFETs often exhibit high 1/f and flicker noise due to charge traps and such in the device so again are often less suitable for low-noise inputs.
As Scott implies, IME noise rise and fall with temperature, especially at audio frequencies, is often due to other noise sources than basic resistive thermal (Johnson) noise.
Years ago a fellow engineer used some wideband (microwave) GaAs devices and then later SiGe HBTs from an RF LNA (low noise amplifier) design of mine to (attempt to) create a low-noise audio circuit. Alas, I had to point out after he built it that the devices he chose had 1/f corners in the hundreds of MHz and thus "low noise" only applied at microwave frequencies... Need the right tool for the job.
Some of my undergrad classes included tube circuits, but I've not designed a tube circuit in many years, so will pass on digging into them again.