Lataa Tenebra 0.6.1

<!-- tenebra-promotion-sha256:0e37bee00a3e5e0948d1c41d5720096d7674f01cce2dac9fb329de15a6dcc058 -->
# Tenebra 0.6.1

This desktop hotfix prevents failures in Tenebra's on-connect,
multi-destination node validation from being reported as an
all-servers-unreachable result. It also makes failures in the server list's
separate direct TCP status checks visible and prevents one incompatible
subscription node from invalidating the shared sing-box configuration.

• The on-connect validation reserves a free contiguous block of loopback ports
instead of relying on a fixed range that another local process may already
own.
• Its temporary SOCKS listeners use fresh credentials, and readiness is
confirmed with an authenticated SOCKS negotiation. A foreign listener can
no longer be accepted as Tenebra's probe merely because it answers on the
expected port.
• The core watches this validation's probe process throughout startup and
measurement. Detected startup failures, bind races and early process exits
now stop validation with a local error rather than returning an
all-servers-unreachable verdict.
• Probe stdout and stderr are retained in a bounded tail for diagnosis and are
scrubbed before being logged or returned through the control interface.
• Imported VLESS URLs with unsupported stream transports, including `xhttp`,
are excluded before the shared probe or tunnel configuration is built.
VLESS nodes using QUIC without its required TLS are excluded for the same
reason, while compatible VLESS QUIC nodes remain supported. Other usable
nodes in the profile stay available.
• Separately, the server list shows whether its direct TCP status checks are in
progress, ready or failed and provides a retry action. An earlier successful
RTT may remain visible while a refresh runs or fails, but it is marked stale
and is not counted or rendered as a current signal-strength result.

## Verification scope

Automated tests cover the on-connect validation's probe ownership, lifecycle
and port races, authenticated readiness, bounded log capture and the bundled
sing-box configuration, builder-level unsupported-transport isolation and
compatible VLESS QUIC, plus the server list's direct TCP states. The affected
profile reproduced the original failure in a retained Windows VM：one `xhttp`
entry made the bundled sing-box reject the shared configuration with `unknown
transport type：xhttp`, blocking otherwise usable nodes. Native installation,
service, ordinary-user UI and tunnel checks of the exact rebuilt signed
candidate remain required before publication.