Previously this was not working due to this error:

```
ruby/3.1.4/lib/ruby/3.1.0/fileutils.rb:637:in `rm_rf': wrong number of arguments (given 2, expected 1) (ArgumentError)
	from extconf.rb:33:in `<main>'
```

Also update the documentation to reflect that --clean is not enabled
by default.

Cleaning is on by default; it's just that prior to #168 the cleaning
was quietly failing:

```
Cleaning files only used during build.
/Users/stanhu/.asdf/installs/ruby/3.1.4/lib/ruby/3.1.0/fileutils.rb:637:in `rm_rf': wrong number of arguments (given 2, expected 1) (ArgumentError)
	from ./extconf.rb:28:in `block in <main>'
	from <internal:dir>:220:in `glob'
	from ./extconf.rb:27:in `glob'
	from ./extconf.rb:27:in `<main>'
make: [clean-ports] Error 1 (ignored)
```

The comment, "C extension is in the root", is no longer accurate. It's
in ext/ which will be added to the LOAD_PATH by bundler or rubygems.

Add generic `get_ctx_flag` getter and `set_ctx_flag` setter

Signed-off-by: Daiki Ueno <[email protected]>

We had some build failures on macOS with this error:

```
configure: Use gpgrt-config as libassuan-config
checking for LIBASSUAN - version >= 2.4.2... yes (3.0.1)
checking LIBASSUAN API version... does not match.  want=2 got=3.
```

This occurs when:

1. libassuan v3.0 is installed.
2. `PKG_CONFIG_PATH` includes the Homebrew path of `/opt/homebrew/lib/pkgconfig`.

https://github.com/gpg/gpgme/blob/5bf3e6d0754a179b316f90a885371af99bc03ea1/m4/libassuan.m4#L32
attempts to run `gpgrt-config libassuan --exists`. This will return
the Homebrew version of libassuan if `PKG_CONFIG_PATH` includes it.

To avoid this build failure, preface `PKG_CONFIG_PATH` with the local
versions of libassuan and libgpg-error.

Relates to #128

See https://github.com/gpg/libgpg-error/blob/master/NEWS

- Bumped `gpgme` from `1.21.0` to `2.0.0`
  - https://github.com/gpg/gpgme/blob/master/NEWS
- Bumped `libassuan` from `2.5.6` to `3.0.2`
  - https://github.com/gpg/libassuan/blob/master/NEWS

Changes I had to account for from `gpgme`:
- Removed trust item functions
  - `gpgme_op_trustlist_start`, `gpgme_op_tristlist_next`, and `gpgme_op_trustlist_end` were removed
- Removed `GPGME_ATTR_*` constants
- Timestamp field type change
  - `gpgme_subkey_t`, `gpgme_key_sig_t`, and `gpgme_new_signature_t` changed from `signed long` to `unsigned long`
- New functions
  - `gpgme_op_random_bytes` and `gpgme_op_random_value`
- New constants
  - `GPGME_RANDOM_MODE_NORMAL`, `GPGME_RANDOM_MODE_ZBASE32`, `GPGME_DECRYPT_LISTONLY`, `GPGME_CREATE_GROUP`

Closes #202

…ation

Change verify() to decrypt() for encrypted+signed data in two failing tests:

The tests were incorrectly calling verify() on data created with
crypto.encrypt(text, :sign => true), which produces encrypted+signed data.
The verify() method expects signed-only data, causing "Bad data" errors.

The decrypt() method properly handles both decryption and signature
verification for encrypted+signed data by calling ctx.decrypt_verify()
internally, then processing the embedded signatures.

This aligns with the existing pattern used in other working tests that
handle signed encrypted data.

Signed-off-by: Daiki Ueno <[email protected]>