stack: Fix a parsing bug when code location is not detected

Add test case that reproduces bug and confirm it's fixed.

Improve another test expectation in stack_test.go.

Fixes maruel#56

vendored v2 stack: Fix a parsing bug when code location is not detected

Bump the vendored and go.mod v2 stack to v2.1.1 to ensure the fix for
issue 56 is in.

Merge back CLI tools from v2.1.0.

Confirms that it makes v1 work with go1.16. With this:

  go get github.com/maruel/panicparse/cmd/pp

does the right thing! It leverages an updated version of cmd/ and
internal/, and then uses stack@v2 instead of stack/ here.

Found a way to hack my way through:
- Copied .github/workflows/test.yaml from v2.1.0.
- Deleted all of vendor/. It was breaking the CI tests.
- Updated stack/context_test.go to make it work with go1.16.
- Updated stack/context_test.go to make it work with all of vendor/
  removed.
- Use v2.1.0 version of the executables in cmd/ and //main.go.
- Use v2.1.0 version of the internal/.
- Copied stack@v2.1.0 back into
  vendor/github.com/maruel/panicparse/v2/stack. Since older Go versions still
  look into it, they are able to load v2 even if go modules are disabled.

github actions: tweak caching

Another try to improve caching. I experimented with my repository msgbus
and this new configuration seems to work better.

restore-keys it doesn't seem to work as expected, it seems to always
cause the step to restore from the default key, which is quite weird.

stack: add simple parseStack example (maruel#51)

internal: Nudge to ugprade to v2

If you are reading this, you have three choices:

- Upgrade to v2. Seriously, do this. It's so much better!
- Disable the nudge with the environment variable PANICPARSE_V1_NO_WARN=1
- Stick to v1.5.0.
I

Fix go get path in README.md and panic help

Using:
  go get github.com/maruel/panicparse/cmd/pp

will retrieve 1.5.0. But now that 2.0.0 is in the sumdb, I can't remove
it anymore, hence a new 2.0.1 release that fix the path in the README
and panic's help.

README.md: add link to v2.0.0 blog post

stack: fix race ToHTML address format to hex

A pointer in decimal isn't particularly useful.

stack: speed up pass through by 1.9x

Use lower level bufio functionality, work in []byte instead of string,
saving a ton of memory copy and making the pass through much faster.

Interestingly parsing became a tad slower with more memory allocation.
It is unclear to me where they occurs.