gh-156505: Speed up difflib.HtmlDiff for lopsided replacements

[felix314159]

difflib._mdiff() used lists as FIFO queues for unmatched lines and removed
items with pop(0). This repeatedly shifted the remaining items, causing
quadratic behavior for lopsided replacement blocks.

Use collections.deque and popleft() for constant-time FIFO removal. This
preserves the existing output while making the affected line-pairing step scale
approximately linearly.

Benchmarks on matched release builds from CPython main:

Case	Before	After	Speedup
_mdiff(), 64,000-to-1	264 ms	61.9 ms	4.26x
HtmlDiff.make_table(), 64,000-to-1	398 ms	206 ms	1.93x
_mdiff(), 128,000-to-1	999 ms	126 ms	7.94x

The complete test_difflib suite passes. A focused regression test covers
lopsided replacements in both directions, and randomized differential testing
confirmed identical output for _mdiff() and HtmlDiff across context and
wrapping modes.

Fixes #156505

• Issue: difflib.HtmlDiff has quadratic line pairing for lopsided replacements #156505

[bedevere-app]

Most changes to Python require a NEWS entry. Add one using the blurb_it web app or the blurb command-line tool.

If this change has little impact on Python users, wait for a maintainer to apply the skip news label instead.

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[felix314159]

I think the skip news label makes sense here as behavior is unchanged

[eendebakpt]

Do you have benchmarks for more mixed scenario's? In particular cases where the changes could cause a regression?

[felix314159]

I benchmarked the parent commit against this PR using 15 paired randomized samples with the same non-debug interpreter. So basically I tested identical inputs, sparse replacements, balanced replacements, and mixed insertion/deletion/replacement workloads. The larger mixed cases ranged from unchanged to like 2% faster. A four-line _mdiff case measured about 1.45% slower, but remember this is only around 0.3 µs per call. The corresponding HtmlDiff case differed by only 0.15%. So IMO there is no meaningful regression outside the lopsided case this PR fixes.

[StanFromIreland]

CC @dg-pb @tim-one

[dg-pb]

Sensible improvement. No new imports. Change makes sense in straight forward manner. No risk for degradation for any cases. Not much to add really.