fix(lamni): zero manual_shift_x/y at the start of a fresh alignment run

LamNI's alignment-start reset (_align_impl(), documented as "the actual
start of a fresh alignment run") already resets reset_correction()/
reset_xray_eye_correction() there, but never touched manual_shift_x/y --
unlike flomni's equivalent, which this session's work already fixed to
zero manual_shift_x/y at the same point. Same risk: a leftover manual
shift from a previous sample would otherwise bias the next sample's
alignment tomogram.

Found via an audit of whether this session's flomni-side changes (+ the
merged fixes/flomni4 branch) called for any analogous LamNI changes --
this was the one confirmed real gap; everything else either already
existed on LamNI, is generic/shared infrastructure that needed no
changes, or is structurally inapplicable (LamNI is 360-degree-only
laminography).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NPZ9gzN2rt3wgiMnTKyrug

csaxs_bec/bec_ipython_client/plugins/LamNI/x_ray_eye_align.py

@@ -336,6 +336,8 @@ class XrayEyeAlign:
         # __init__ for why this can't live there.
         self.lamni.reset_correction()
         self.lamni.reset_xray_eye_correction()
+        self.lamni.manual_shift_x = 0
+        self.lamni.manual_shift_y = 0
 
         self.gui.enable_submit_button(False)