Update to toolchain 3.1.0-gcc10.3

[earlephilhower]

No description provided.

[d-a-v]

LGTM !
Flash occupation for a random ~500KB flash sketch was reduced by ~500 bytes

[earlephilhower]

Can you try again? BearSSL was rebuilt using one of the new optimization settings.

We also need to try the generated code out. I'm not 100% confident of all the changes at this point. There were a lot of machine definition changes...

[d-a-v]

I just retried w/ BSSL/MFLN test.
It is working and ~1700 bytes smaller.
👍🏽
(inet issue: I replaced tls.mbed.org by www.github.com)

[jjsuwa-sys3175]

the patches that seem to have remarkable effects with this update IMHO:

• "Defer storing integer constants into litpool until reload"
  until now, early pooling of constants hindered integer optimizations.
• "Add support for sibling call optimization"
  it reduces funccall overhead and stack usage.
• "Improve constant synthesis for both integer and floating-point"
  new synthesis patterns, and early synthesis hindered integer optimizations until now.
• and a couple of mods related to logical AND with constants...

(note: fixed-size/value aligned memset() goes inlined small loop especially on ESP32 due to zero overhead looping)

[earlephilhower]

Thanks @d-a-v ! I'd suggest we merge this ASAP to get as many user hours on it as possible before a release now that some complicated tests are proven good...