runtime: make runtime panics recover()-able

[dgryski]

No description provided.

[aykevl]

Did you check the binary size impact?
In any case, this results in an unnecessary heap allocation in a very sensitive area (for example, when an "out of memory" runtime panic is triggered, allocating would lead to infinite recursion). So I suggest factoring out the recover part of runtime._panic to avoid the conversion to an interface.

(I'm a bit careful with this change, as I think there might be reasons why it's the way it is now. Just want to make sure we don't overlook anything).

[dgryski]

No, I didn't. I also noticed that CI failed. Marking as draft for now.

[aykevl]

Actually, the fact that CI fails is because of this heap allocation:

/Users/runner/work/tinygo/tinygo/build/tinygo build -size short -o test.hex -target=pca10040 -gc=none -scheduler=none examples/blinky1
tinygo:ld.lld: error: undefined symbol: runtime.alloc
>>> referenced by string.go:62 (/Users/runner/work/tinygo/tinygo/src/runtime/string.go:62)
>>>               /Users/runner/Library/Caches/tinygo/thinlto/llvmcache-059CC3AC26AD6D48C9F8BD6A8548A01527722C0E:(runtime.stringConcat)

[dgryski]

This is needed to make the bytes package from 1.23 pass tests.

[aykevl]

I have made a new PR with a somewhat different approach: #4758