Fix connect-time username prompt never reaching SSH/RDP authentication

The username entered at the connect-time prompt (added for issue #21)
was updating SessionTab's own in-memory Profile copy, but
SshSessionBackend/RdpSessionBackend are constructed with -- and only
ever read from -- their own separate Profile copy on a worker thread,
which never saw that edit. Authentication was still built from the
original (blank) username regardless of what was typed into the
prompt.

SessionConnectOptions gains a username field, populated by SessionTab
on every connect attempt and threaded through the same way password
already is; both backends now prefer options.username over
profile().username. Covered by a new SSH regression test using an
exact-match fixture host that only succeeds for a specific
user@host target.

Bump version to v2026.9.16.5.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs/PROGRESS.md

@@ -193,6 +193,22 @@ Delivered:
   told the user to go edit the profile instead of ever prompting inline.
   Removed; connect-time prompting is now the only username gate for
   SSH/RDP
+- Even with the three checks above gone, a username entered at the
+  connect-time prompt still never actually reached SSH or RDP
+  authentication: `SshSessionBackend`/`RdpSessionBackend` are constructed
+  with their own `Profile` copy up front (moved to a worker thread) and
+  read `profile().username` directly, which never sees `SessionTab`'s
+  later edit to its own in-memory profile once the user answers the
+  prompt. `SessionConnectOptions` (which already carries `password` the
+  same way) gained a `username` field, populated by `SessionTab` from
+  its profile copy on every connect attempt; both backends now prefer
+  `options.username` over `profile().username` when building the actual
+  connect target/auth call. Covered by a new SSH regression test
+  (`tests/fixtures/fake_ssh.sh`'s `requireuser` host only accepts an
+  exact `prompted-user@requireuser` target, so the test fails unless the
+  option, not the stale profile copy, is actually used) -- RDP has no
+  equivalent fake-server test harness, so that side relies on mirroring
+  the already-tested `m_activeOptions.password` pattern exactly
 - Robustness fix: an unrecognized `FramebufferUpdate` rectangle encoding
   used to abort the connection generically; `kAnnouncedEncodings` is now
   the single source of truth for what `SetEncodings` announces and what