Add M2a plan: thirteen tasks, tests, fake server and recordings

The tasks build the inference path: emsha-backed SHA-256, inferproxy,
config, a hand-written HTTP and SSE client, request building, delta
assembly, the chat state machine, the thinking cap, the slot gate with
retry, the startup self-test and on-device verification.

Everything the tasks copy in was checked against a private reference
implementation: the gate passes after each task in order, the timing
tests pass repeatedly under CPU load, and the reference passes the
self-test and all four device checks on straylight. Expected results
for the recorded streams were derived by a separate script.

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

docs/plans/M2a/files/crates/loopd/tests/fixtures/config/full.toml

@@ -0,0 +1,31 @@
+[infer]
+socket = "/tmp/infer.sock"
+model = "some-model"
+
+[slots]
+main = 2
+background = 3
+
+[expect]
+template_sha256 = "000102030405060708090a0b0c0d0e0f101112131415161718191a1b1c1d1e1f"
+n_ctx = 4096
+slots = 4
+
+[sampling]
+temperature = 0.2
+top_p = 0.9
+top_k = 40
+
+[limits]
+poll_ms = 50
+busy_wait_ms = 200
+load_wait_ms = 300
+idle_grace_ms = 150
+liveness_ms = 100
+thinking_cap = 20
+thinking_overrun = 10
+max_tokens = 512
+queue_len = 1
+retry_attempts = 2
+retry_backoff_ms = [10]
+retry_window_ms = 1000

docs/plans/M2a/files/crates/loopd/tests/fixtures/config/minimal.toml

@@ -0,0 +1,12 @@
+[infer]
+socket = "/run/boxmaker/infer/infer.sock"
+model = "ornith-1.5-35b-a3b"
+
+[slots]
+main = 0
+background = 1
+
+[expect]
+template_sha256 = "f55f52930aa8bf44ab5cb85f99370fcc3c56e9a85640b812086d5330bce5d86b"
+n_ctx = 131072
+slots = 2