Add DELTA and DELTADELTA transform codecs

[xiangfu0]

Context

Stacked on #19285 and part of the split of #18229.

What changed

• Adds DELTA and DELTADELTA as typed-layout-preserving INT/LONG transforms.
• Keeps the output headerless and the same width/count as the input so transforms can be chained before packing or compression.
• Defines intentional two's-complement modular arithmetic for overflow-adjacent values.
• Registers both frozen codec names in the closed runtime registry.
• Makes persisted decoding explicitly big-endian, independent of a caller view's byte order.

Safety and scope

Registering DELTA/DELTADELTA makes them available to the bounded runtime API, but production forward-index creation/loading is not wired until #19307. This slice changes no segment format or production I/O path.

Verification

• INT/LONG round trips for empty, boundary, monotonic, negative, sign-changing, and overflow-adjacent data.
• Allocating and bounded decodeInto paths, transform chains, and malformed-width rejection.
• Stable hand-authored decode fixtures that pin DELTA and DELTADELTA wire bytes independently of the encoders.
• Little-endian caller-view regressions proving big-endian persisted decoding.
• Spotless, Checkstyle, license format, and license check for pinot-segment-local.

Stack

#19284 → #19285 → #19305 (this PR) → #19306 → #19307 → #19308 → #19309

Review and merge parent-first.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 96.58537% with 7 lines in your changes missing coverage. Please review.
✅ Project coverage is 67.21%. Comparing base (4824a5f) to head (b81db76).

Files with missing lines	Patch %	Lines
...gment/local/io/codec/BaseDeltaCodecDefinition.java	87.80%	3 Missing and 2 partials ⚠️
...ment/local/io/codec/DeltaDeltaCodecDefinition.java	97.97%	1 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             master   #19305    +/-   ##
==========================================
  Coverage     67.20%   67.21%            
- Complexity     1418     1424     +6     
==========================================
  Files          3479     3482     +3     
  Lines        223075   223280   +205     
  Branches      35135    35166    +31     
==========================================
+ Hits         149920   150080   +160     
- Misses        61184    61255    +71     
+ Partials      11971    11945    -26     

Flag	Coverage Δ
integration	100.00% <ø> (+100.00%)	⬆️
integration1	100.00% <ø> (?)
integration2	0.00% <ø> (ø)
java-25	67.21% <96.58%> (+<0.01%)	⬆️
lane-a	100.00% <ø> (+100.00%)	⬆️
lane-b	0.00% <ø> (ø)
temurin	67.21% <96.58%> (+<0.01%)	⬆️
unittests	67.21% <96.58%> (+<0.01%)	⬆️
unittests1	57.64% <0.00%> (-0.06%)	⬇️
unittests2	39.01% <96.58%> (+0.01%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

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

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[Jackie-Jiang]

Do we need DELTADELTA, or just chain 2 DELTA?
What is the pros and cons of both approaches?

[xiangfu0]

They are close but not interchangeable, so we kept both — DELTA,DELTA stays legal in the DSL (the validator accepts chained value-preserving transforms), while DELTADELTA is the dedicated form. Trade-offs:

Not the same encoding at index 1. DELTADELTA emits [v0, v1−v0, second differences…] (the classic Gorilla-style form). Composing DELTA twice emits [v0, (v1−v0)−v0, second differences…] — the second slot carries an epoch-magnitude outlier for timestamp columns. Both round-trip exactly, but when a frame-of-reference packing stage follows (DELTADELTA,T64,…), that one outlier forces max bit-width for the whole first 64-value block; the dedicated codec keeps every post-v0 value small.

Cost per stage. The executor runs each stage as a full pass with its own intermediate buffer, so DELTA,DELTA is two passes + one extra scratch buffer per chunk on encode and decode; DELTADELTA does the second difference in a single pass.

Frozen-name surface (the real argument for dropping it). Codec names are permanent on-disk contracts once V7 headers ship. Dropping DELTADELTA later would be a format break, whereas adding it later would have been compatible — so keeping it is a commitment. We think it earns it as the canonical transform for the headline use case (near-regular timestamps), but if you prefer to minimize the frozen surface, removing it from the registry in this PR and re-adding it later is fully backward-compatible — happy to do that instead.

🤖 Addressed by Claude Code