No, I don't have a direct reference.
Generally, this belongs to a group of issues which stem from the SoC character of the STM32, i.e. that they are assembled from modules. Consequence for documentation is, that the relationship between modules is poorly described in the RM (this demonstrates itself in many ways, e.g. incosistent signals naming, poor or no description of some of the interconnects, "Interconnections" chapter in RM is an afterthought missing for older families, etc.).
In this particular case, the thing is, that modules which control GPIO (in their AF mode) usually control in some way also the output-enable signal (i.e. the one which makes the pin Hi-Z). This is especially so for peripheral ports, which can be "turned" - in this case, all TIMx_CHx can be "turned" i.e. used either as input (if you set given channel as Input Capture) or output (if it's set as Output Compare). Input Capture essentially means, that the pin has to be Hi-Z, so the control is already there. And it is easy to test, that in Output Compare setting it *is* controlled by the TIMx_CCER.CCxE bit (and/or CCxNE, where applicable).
There's another indication for this, too, but not in the 'L0xx RMs - for Advanced Timers (and there's none in 'L0xx) there's a Output control bits for complementary OCx and OCxN channels with break feature table, from which it's quite obvious that cleared CCxE/CCxNE means HiZ AF pin.
---
> Also, the PWM signals all have a pull-down.
External? And connected so that it ensures the LED will be OFF?
Well, at any case, in debugger, stop the mcu when LED is in the unexpected 1 state, and measure + read out relevant GPIO and TIM registers content and check/post.
JW
