In the ESP32 blink demo, what are the roles of the laptop, the cross-compiler and FreeRTOS?

The laptop (Linux) writes the C app and builds it with a cross-compiler that produces machine code for the ESP32's Xtensa processor; the image, containing the app and FreeRTOS, is uploaded over USB; on the board FreeRTOS schedules the task that switches the LED on and off.

The demo splits the work between two machines:

• Laptop – Linux OS. Where you want to program the LED. It holds the C source, the FreeRTOS headers and libraries, and Espressif's toolchain, including a cross-compiler.
• Build & upload. The cross-compiler runs on the laptop's x86 processor but emits code for the ESP32's Xtensa processor. The result is one firmware image, your app linked with FreeRTOS. It is flashed over the USB-UART bridge.
• IoT device – ESP32 + FreeRTOS. Where the timing decision is made: FreeRTOS runs the blink task, which turns the LED on and off at a regular programmed interval.

A cross-compiler is needed because the target is a different architecture from the build machine, and it is far too small to run a compiler itself.

The LED itself needs no bus protocol such as I2C: it hangs off a plain GPIO pin (IO5) that the task switches high and low.

Go deeper:

• Wikipedia — Cross compiler — why embedded builds use host and target toolchains, and the "Canadian cross".
• ESP-IDF — Blink example — Espressif's official version of this demo, ready to build.