esp-idf-config/sdkconfig.defaults: Fix TLS access to api.github.com; remove BLE from 4MB ESP32-S3

[dhalbert]

• Fixes Unable to connect to api.github.com #8910
• Fixes Slower https access on 9.0.0 #8919

Both of the options below were not set (and therefore not turned on), and it caused TLS access to api.github.com to not work at all.

# CONFIG_MBEDTLS_ECP_NIST_OPTIM is not set
# CONFIG_MBEDTLS_ECP_FIXED_POINT_OPTIM is not set

Turning on either one fixes the problem. Turning on CONFIG_MBEDTLS_ECP_NIST_OPTIM instead of CONFIG_MBEDTLS_ECP_FIXED_POINT_OPTIM is smaller (+2.2kB vs +6.9kB) and faster (1.6 seconds vs 7.6 seconds, at least some of the time).

I don't understand why some "OPTIM" must be turned on for it to work at all, but I'll accept that at face value. I experimented with various combinations.

tag @justmobilize

EDIT: This didn't fit on a few boards, so I went ahead and did the refactoring and BLE trimming we talked about:

• Turn off _bleio by default on ESP32-S3 4MB boards.
• Turn off _bleio by default in mpconfigport.mk for C3 and C6, and then removed that setting from a bunch of mpconfigboard.mk files.
• Turn on gifio and jpegio on any boards with displayio enabled (consulted with @jepler about this). Previous those modules were dependent on espcamera, which may just have been copy pasta or something like that.

[justmobilize]

I tested this with:

• Adafruit MagTag with ESP32S2 - passed
• FeatherS3 with ESP32S3 - passed
• Adafruit Feather ESP32-S3 TFT with ESP32S3 - passed

[jepler]

so the theory is, github gets tired of waiting for key exchange without the optimized implementation of the algorithm?

[dhalbert]

so the theory is, github gets tired of waiting for key exchange without the optimized implementation of the algorithm?

That could be. I looked at the code and I don't see that's it non-functional without the optimizations.

I think we want the optimizations anyway because it's quite slow without them in 9.0.0 compared with 8.2.x, enough to provoke #8919.