Add Plot2D.set_display_window — re-window contrast without re-quantising

[CSSFrancis]

Why

set_clim re-encodes the cached raw frame over the new range. That is the right
default — the codes then span exactly the visible band, so the contrast on screen
gets all 8 bits — but it means the pixels move, and the previous band is gone:
everything outside the new range is saturated to 0/255 in the codes that get
stored.

For a serialised figure that is the difference between having a contrast
control and not having one. build_standalone_html writes the codes plus
raw_min/raw_max, and the JS rebuilds its LUT from display_min/display_max
over that band (_buildLut32). So a saved page can re-window freely inside the
band it was encoded with, and not at all outside it. Encode with set_clim at the
display window and the band is the window: an identity LUT, nothing left to
move.

What

The non-destructive path already existed — inlined in set_clim's tile branch,
where re-quantising would re-encode a full-res frame on every drag tick. This
promotes it to a documented public method so a caller can choose the trade
deliberately:

p = ax.imshow(frame, vmin=lo, vmax=hi)   # quantise over a WIDE band
p.set_display_window(black, white)       # window inside it, no re-encode

Nothing is re-encoded and nothing travels but two floats. The cost is precision —
a window much narrower than the encoding band resolves in coarser steps — so the
docstring says to quantise over the range you want to be able to reach, and
set_clim gains a cross-reference so the pair is discoverable from either side.

No behaviour changes for existing callers.

Tests

anyplotlib/tests/test_plot2d/test_display_window.py, 10 cases pinning the
distinction from both sides:

• set_display_window leaves image_b64 and the raw_* band untouched;
• set_clim re-encodes and collapses raw_* onto the window;
• a wide band keeps room to window in both directions, and set_clim first
  would have thrown that away;
• set_clim in tile mode and set_display_window produce identical state, so
  the two cannot quietly diverge.

Full suite: 2122 passed, 58 skipped.

Context

Written for SpyDE's report export, where a reader opening a saved HTML report
could not adjust the contrast of a figure at all — the pixels arrive hard-clipped
to whatever window was chosen when the report was written.

Assisted-by: Claude Opus 5 (1M context)

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.88%. Comparing base (729cb16) to head (4ef7b68).
⚠️ Report is 4 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #61      +/-   ##
==========================================
+ Coverage   90.74%   90.88%   +0.14%     
==========================================
  Files          40       40              
  Lines        4526     4532       +6     
==========================================
+ Hits         4107     4119      +12     
+ Misses        419      413       -6     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.