fix: Don't use hidden progress bar

[lcian]

Description

• Instead of using a hidden indicatif progress bar when SENTRY_NO_PROGRESS_BAR, use no progress bar instead, and just forward messages on it to log::debug/info.
• Also, use std::io::stderr().is_terminal() to determine whether or not to use a progress bar when SENTRY_NO_PROGRESS_BAR is not specified.
• This prevents the progress bar from "swallowing" messages that are set on it AND also those logged using log, as our log utility ultimately forwards to the progress bar when one is active.

Issues

Close #2829
Close CLI-180
Close #2790
Close CLI-171

Testing

1. sentry-cli debug-files bundle-jvm --log-level=debug

stdout and stderr before and after the change:
stdout_before.txt
stdout_after.txt
stderr_before.txt
stderr_after.txt

The stdout remains the same.
The stderr changes, now we see more output, such as the file being currently processed (DEBUG level), and other messages at INFO level, such as Skipping ./File.java because it is not valid UTF-8 which is what motivated this PR in the first place.

2. sentry-cli sourcemaps upload --log-level=debug

stdout and stderr before and after the change:
stdout_before.txt
stdout_after.txt
stderr_before.txt
stderr_after.txt

The stdout remains the same as well.
The stderr now shows 2 POST calls to /artifactbundle/assemble, in contrast with the single POST call to that endpoint that we would see without the change.

[linear]

CLI-180 Avoid swallowing of messages when using progress bar and not running in terminal

[lcian]

Without this PR we're indeed missing not only the messages emitted on the progress bar itself (with pb.set_message) but also those emitted with log:: macros. This is because here we use the progress bar to print those messages too, if one is active:
see https://github.com/getsentry/sentry-cli/blob/master/src/utils/logging.rs#L92
So this will solve that problem too

[linear]

CLI-171 Logs are missing from `sourcemaps upload`

[szokeasaurusrex]

Amazing find, thanks so much for the PR!

I added some minor suggestions for the implementation; overall, this is an excellent improvement though!

Side note: it is interesting that we do two assemble endpoints for the sourcemap upload in the "after" logs, as the first assemble request already shows all the chunks are uploaded, the second call is redundant. But, I suppose this is an edge case, so it is fine, but it is interesting behavior that we would not have observed without your change

[szokeasaurusrex]

lgtm, besides one comment on the changelog

[szokeasaurusrex]

🚀