It erroneously applied another layer of MIME header encoding.
The problem is that existing Sieve scripts may rely on this behaviour; i.e. scripts may contain addresesses with literal UTF8 in the phrase part.
Therefore, this change allows both behaviors.

lib-sieve: util: rfc2822: Added rfc2822_header_write_address() which allows adding a header with one or more addresses.

It applies UTF8 encoding only when the address string somehow contains 8-bit characters.
Sieve addresses are supposed to have any UTF8 characters in the phase MIME-encoded, not literal 8-bit characters.
However, this has been allowed from the beginning, so disabling the old behavior may break existing installations.
This function allows supporting both options.

Also dropped the single-letter identifier for most fields to prevent conflicts with mail user variables; these were undocumented anyway.

A few were still omitted.

Cleanup performed with the following semantic patch:

	@@
	expression E;
	@@

	- if (E != NULL) {
	- 	i_stream_unref(&E);
	- }
	+ i_stream_unref(&E);

	@@
	expression E;
	@@

	- if (E != NULL) {
	- 	o_stream_unref(&E);
	- }
	+ o_stream_unref(&E);

Cleanup performed with the following semantic patch:

	@@
	expression E;
	@@

	- if (E != NULL) {
	- 	timeout_remove(&E);
	- }
	+ timeout_remove(&E);

Cleanup performed with the following semantic patch:

	@@
	expression E;
	@@

	- if (E != NULL) {
	- 	io_remove(&E);
	- }
	+ io_remove(&E);

	@@
	expression E;
	@@

	- if (E != NULL) {
	- 	io_remove_closed(&E);
	- }
	+ io_remove_closed(&E);

Error handling is delayed, so that we need to explicitly handle the error in Sieve.
This is now implemented by handling the error before executing the scritpt and placing the parsed address in struct sieve_script_env.

lib-sieve: Added sieve_script_env_init() to initialize struct sieve_script_env from struct mail_user.

Most code was put in an unnecessary else block.

If a home directory is given, it's assumed that it will be autocreated even
if it doesn't initially exist. Earlier the autocreation may have been done
by lib-storage as it created the mail root directory, but this no longer
happens.

The parent event doesn't really matter here, so NULL is used.

This follows the following commits in Dovecot:
lib-storage: Moved connection information in struct mail_user into separate struct mail_user_connection_data.
lib-storage: mail-user: Added more information about the client connection.

lib-sieve: vnd.dovecot.report extension: Fixed creation of "Original-Mail-From" and original "Original-Rcpt-To" headers.

Recent changes relating to lib-smtp caused the paths to have duplicate '<' '>'.

Now using struct smtp_address rather than sieve_address.
In some cases, struct message_address may prove to be more suitable, but this is the best fit for now.
Adjusted to the renamed fields in struct mail_deliver_context.

It doesn't serve a purpose either, since parsing "from" address from URI is not allowed for notify mailto.

lib-sieve: store action: Avoid NULL pointer dereference during rollback when transaction context is somehow unassigned.

This change matches the other transaction stages.
This will not currently happen though.
Reported by Coverity.

Otherwise the error status is overwritten.
Problem found by Coverity.

Used the stream's cur_header field, rather than the local variable.
Problem found by Coverity.

The merge_from_parent() function may not actually read anything, potentially leaving the stream buffer unassigned.
Problem found by Coverity.

This makes the implementation easier to understand and amend once Dovecot APIs changes.
This implicitly makes the implementation suitable for the new istream snapshot handling.

Added assertions on subtractions.
Restructured the code for clarity.
Added comments.

Ported Dovecot's istream-jsonstr for ManageSieve syntax.

Call master_login_init() before master_service_init_finish(), which frees
all the data stack done in initialization.

This didn't normally cause any visible problems, because data stack wasn't
currently being used in a way that the strings were invalidated. However,
it was causing failures if --enable-devel-checks was used.

Patch by Andreas Schulze.