fix: correctly read execution args from watch command (#10381)
### Description
This is a follow-up fix to:
https://github.com/vercel/turborepo/pull/10236
As it turns out, this change presented a regression in how the
`--concurrency` flag behaves in watch mode. Previously, since this flag
was read from the `execution_args` directly, it worked as expected.
After switching it to read from `config`, it no longer worked in `watch`
mode, as `Args` would only extract correctly from the `Run` command.
This change lets the `Args::execution_args()` getter return the
execution args from the watch command that has been run, if present.
This _is_ a pretty significant change, as previously _none_ of the run
mode execution args would be respected in watch mode. I'm not totally
confident this is desired.
The alternative is to bring back the special case for `concurrency`, and
try to read from execution_args directly when parsing it, and falling
back to `config` otherwise.
### Testing Instructions
1. Clone `main` and `pnpm build:turbo`
2. Run `./path/to/turbo watch <some-cmd>` in a repository where more
than 10 tasks (the default) are persistent
3. Observe that even specifying `--concurrency=<task-count>` does not
override this setting
4. Checkout the v2.5.0 commit and rebuild
5. Run the same test, and observe that the concurrency flag works as
expected
6. Finally, clone this branch and rebuild
7. Run the same test, and observe that the concurrency flag works as
expected
