Drop attachment captions

[Bilb]

Opened against the attachment-availability branch of this fork, so the diff is exactly the one commit this PR is about. Merges after that branch does; it targets client in session-foundation as #168.

caption on an attachment was a pure pass-through with no consumer, so this removes it end to end.

What goes

• OutgoingAttachment::caption and Attachment::caption (include/session/client/attachment.hpp)
• the caption TEXT column on message_attachments in full_schema.sql, plus a new migration src/client/schema/006_drop_attachment_caption.sql dropping it from existing databases
• the four statements in src/client/client.cpp that carried it: the attachment read, the outgoing write, the outgoing protobuf build, and the incoming write
• optional string caption = 11 in AttachmentPointer
• the incidental assertions in tests/test_client/attachments.cpp (the tests shrink; none are deleted)

Why it is safe

Nothing reads it. It went protobuf in -> column -> struct field -> column -> protobuf out and nowhere else: no client renders it, nothing searches on it, and it has no index, default or foreign key. The conversation-preview query in client.cpp already excludes it on purpose and says so in a comment.

Removing it also takes four positional statements down by one column each, which is four fewer places for a template parameter list, a ? count and a structured binding to drift apart.

Field 11 is reserved, not freed

AttachmentPointer gains its first reserved block:

reserved 11;
reserved "caption";

Old clients on the network keep sending field 11, so the number must never be reused - anything assigned to it later would parse their captions as its own data. This PR deliberately does not touch any other field number.

Protobuf regeneration

proto/SessionProtos.pb.h, proto/SessionProtos.pb.cc and proto/debug_print.cpp are checked in and are not generated at build time, so they are regenerated here with protoc 3.21.12, the version that produced the existing committed files (WebSocketResources.pb.* regenerates byte-identical, confirming the match). Outside the caption accessors the only churn is the expected _has_bits_ renumbering for the fields after 11.

Verification

On top of session-foundation#168 (a8a81514):

• full build clean
• testAll green: 405 test cases, 25,758,663 assertions, including all 47 [attachments] cases
• tests/schema_history_check.sh walks this commit, Attachment availability: say what we hold of a file, and tell everyone showing it session-foundation/libsession-util#168's tip and the current client tip, and all three upgrade cleanly through the new migration

Each of the four positional statements was re-derived against the post-session-foundation#168 code rather than replayed, since session-foundation#168 added unavailable and cached to message_attachments and changed the load query. Recounted by hand afterwards: the attachment read is 12 columns / 12 bindings / 12 template parameters, the outgoing write 8 / 8 / 8, the outgoing protobuf build 8 / 8 / 8, and the incoming write 12 / 12 / 12.

One pre-existing failure further back in the schema-history walk is untouched by this branch: a database from 3df4ae52 (v1.8.0-408) does not reach today's schema because of the group-member state CHECK constraints. It reproduces identically on a pristine client build, so it is unrelated to this change and to session-foundation#168.