Add bundled_proj_sqlite feature to support windows-msvc

[yutannihilation]

• I agree to follow the project's code of conduct.
• I added an entry to the project's change log file if knowledge of this change could be valuable to users.
  • Usually called CHANGES.md or CHANGELOG.md
  • Prefix changelog entries for breaking changes with "BREAKING: "

---

In order to make proj-sys work on Windows (more specifically, windows-msvc. I don't know well about windows-gnu), I found there are at least two options.

1. Use vcpkg to install PROJ
2. Bundle PROJ including SQLite

(I don't describe the details about first option, but here's what it looks like: MIERUNE/point-tiler#64)

For the second option, the proj crate and proj-sys crate already exposes bundled_proj, but I found it's not enough on Windows' case, where we usually doesn't have SQLite installed. So, this pull request adds a feature flag bundled_proj_sqlite to expose the libsqlite3-sys's bundled feature in order to bundle SQLite as well.

That's all of this pull request, but I have to explain about some tricky things:

• PROJ uses the SQLite executable as well as the library (cf. https://proj.org/en/stable/install.html#build-requirements). So, even if we use bundled_proj_sqlite, we need to install SQLite on the GHA CI.
• For some reason, cargo test requires some Windows system libraries, otherwise it fails with the following error. So, I added these in build.rs.

  = note: libproj_sys-119818f7f11b3538.rlib(filemanager.obj) : error LNK2019: unresolved external symbol __imp_CoTaskMemFree referenced in function proj_context_get_user_writable_directory␍
          libproj_sys-119818f7f11b3538.rlib(filemanager.obj) : error LNK2019: unresolved external symbol SHGetKnownFolderPath referenced in function proj_context_get_user_writable_directory␍
  

• The proj::test::test_inverse_projection test reliably fails with STATUS_ACCESS_VIOLATION. I'm yet to figure out what this error is. (fixed now)

[yutannihilation]

If I skip the test case, I still see STATUS_ACCESS_VIOLATION.

If I run the test case solely, it runs normally (and fails with a normal error).

❯ cargo test -- test_inverse_projection
    Finished `test` profile [unoptimized + debuginfo] target(s) in 0.05s
     Running unittests src\lib.rs (target\debug\deps\proj-e76bc29b00e44a55.exe)

running 1 test
test proj::test::test_inverse_projection ... FAILED

failures:

---- proj::test::test_inverse_projection stdout ----

thread 'proj::test::test_inverse_projection' (38996) panicked at src\proj.rs:1768:9:
assert_relative_eq!(t.y(), 0.8028510000110507)

    left  = 0.8028510000110513
    right = 0.8028510000110507


note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace


failures:
    proj::test::test_inverse_projection

test result: FAILED. 0 passed; 1 failed; 0 ignored; 0 measured; 25 filtered out; finished in 0.01s

error: test failed, to rerun pass `--lib`

[yutannihilation]

It seems above was a real bug that affects all platforms, not only Windows.I pushed the fix, but I'm happy to create a separate pull request.

[urschrei]

It seems above was a real bug that affects all platforms, not only Windows.I pushed the fix, but I'm happy to create a separate pull request.

I think pulling it out would be useful. It might be useful to combine it with the ASAN PR?

[yutannihilation]

Thanks, I removed the commit from this pull request and pushed to #246

[michaelkirk]

I don't personally have a way to test this, but it seems reasonable. Any concerns @urschrei?

[urschrei]

Same re: testing, but I'm happy! Sorry we let this slide @yutannihilation.

[yutannihilation]

No hurry at all! I understand it's tricky to review the code that works on platforms that you don't use. Please let me know if there's anything I can help.