0.3.66: adopt ciris-server 0.5.203 — a split node signs its consent as itself

[emooreatx]

The pin moves v0.5.202 -> v0.5.203; the substrate is unchanged, so the lock
moves exactly the server and lens-core entries. 0.5.203 fixes CIRISServer#563
(agent-hosted split-key nodes: the consent guard, the held node pen, the
served key record). This node is standalone — its configured key IS the
node — so none of that path runs here; the pin moves so the two nodes on the
canonical host run one server build.

clippy -D warnings clean, 114 tests pass, cargo metadata --locked proves
the lock (version bumped before the refresh), one copy of each crate.

🤖 Generated with Claude Code

[emooreatx]

CI note, not a pin problem: clippy + build + test failed on diag::tests::live_bytes_track_a_real_allocation —

holding 32MB should not shrink the live figure (10957504 -> 10859008)

That is the mallinfo2 flake the server hit on the same test in its own diag.rs during 0.5.200: a 32 MiB Vec is served by mmap, so it never lands in uordblks, and another test thread's free between the two reads shrinks the figure. The server's fix (CIRISServer#554) reads uordblks + hblkhd and holds a 64 MiB block. This crate's copy still reads uordblks alone, so the same flake will recur here at random — one for the status side to port; I am not editing this crate's diag.rs from the pin PR. Rerunning the job; 114/114 pass locally and passed on #77 twice.