Remove benchmark stats from README, keep feature matrices (#2287)

[brendancol]

Summary

• Drops the benchmark timing tables from README.md so the document focuses on what features ship and where they run.
• Removed: read perf (real-world + synthetic), write perf, reproject perf, full-pipeline timings, merge perf, and the "5-7x faster than rioxarray" line in the GDAL notes section.
• Kept: feature matrices with backend support markers, the projection support matrix, algorithm and source citations, and the pixel-exact consistency note (correctness claim, not a benchmark).

Benchmarks fit better with the suite that produces them, not the feature landing page.

Closes #2287

Test plan

• README.md renders cleanly on GitHub (preview)
• No remaining timing tables or "Xx faster" claims in README
• Feature matrices and projection support matrix still intact

[brendancol]

PR Review: Remove benchmark stats from README, keep feature matrices (#2287)

Blockers (must fix before merge)

None. Clean docs-only deletion.

Suggestions (should fix, not blocking)

• README.md:242 -- The line "Same-CRS tiles skip reprojection entirely and are placed by direct coordinate alignment." was previously the closing context for the removed Merge performance table. Without that table, it now dangles at the end of the Reproject/Merge/Resample block. Options: (a) leave it, since it is still correct behavior context for the Merge row; (b) move it into the Merge description in the feature matrix so the detail stays attached to the feature; (c) drop it. Mild preference for (b).

Nits (optional improvements)

• README.md:209 -- The "Consistency:" line previously had a blank line separating it from the write-perf table. After the deletion it sits directly below the "Features" bullet list. Rendered output is still fine because line 208 is empty, so no action needed.

What looks good

• Tight, focused deletion. Only timing tables and the "5-7x faster" claim are gone.
• Feature matrices, projection support matrix, and the pixel-exact consistency note are preserved.
• The "5-7x faster than rioxarray" claim in the GDAL notes section was stripped while keeping the surrounding factual pixel-exact-match context.
• Diff is tight: -59 +1.

Checklist

• No timing tables remain
• No "Xx faster" claims remain
• Feature matrices intact
• Projection support matrix intact
• Pixel-exact consistency note kept (correctness, not benchmark)
• Orphan "Same-CRS tiles" sentence -- see Suggestions
• [n/a] Algorithm correctness, backend parity, NaN handling, test coverage, docstrings -- docs-only change

[brendancol]

Follow-up review

Addressed the one Suggestion from the previous review:

• Fixed -- README.md:218 (the Merge row in the Reproject/Merge/Resample matrix) now carries the "Same-CRS tiles skip reprojection and are placed by direct coordinate alignment" detail in its description column. The orphan sentence at the bottom of the section has been removed (commit f705f4d).

Nit from the previous review was already no-action.

No new findings. PR is ready.