docs: put the production numbers in the upgrade note (#107)

The deaf-node paragraph cited the bench A/B -- two nodes on one machine for half an hour. Both mining machines have since run 1.2.13 in production: zero spurious disconnections against the 741 the unpatched side logged, and 13 and 18 peers held where the unpatched node held 11, ten of them inbound. Also says what the pre-1.2.11 socket pile cost and what returns the ports.

1 files changed, 6 insertions, 3 deletions

diff --git a/README.md b/README.md
index 787f58d..f41c866 100644
--- a/README.md
+++ b/README.md
@@ -67,11 +67,14 @@ Two more reasons, both measured rather than theorised:
 - **Before 1.2.13** a node closed a disconnected peer's socket twice and left the
   closed handle in its `select()` set, where it made `select()` fail on every
   iteration. The node then spent its socket loop in an error path instead of
-  reading its peers. Two nodes side by side on the same machine, same network:
-  the one without the fix held 11 peers, the one with it held 21.
+  reading its peers — the deaf-node behaviour reported since 1.2.7. Two nodes
+  side by side on one machine, same network: the one without the fix held 11
+  peers and logged 741 spurious disconnections; the one with it held 21 and
+  logged none. On 1.2.13 in production, two mining nodes hold 13 and 18 peers
+  with zero.
 - **Before 1.2.11** a long-running Windows node accumulated sockets it never
   released — 1262 of them in 26 hours on one machine, each holding an ephemeral
-  port.
+  port. Restarting returned them; upgrading stops them accumulating.
 
 Your wallet and chain data live in `%APPDATA%\Bitflash` (Windows) or
 `~/.bitflash` (Linux) and are shared by every version, so upgrading is just