[C++][Feature] Add ALP float/double codec with SIMD

[ColinLeeo]

Summary

Add an opt-in ALP encoding for FLOAT and DOUBLE to the C++ implementation,
and expose the new encoding through the Python API.

The existing Gorilla/Chimp codecs are not a good fit for fixed-precision
decimal-like training data. ALP converts such values to integers, applies
frame-of-reference and bit packing, and stores exceptions for values that
cannot be represented exactly. This PR adds:

• TSEncoding::ALP = 15, supported only for FLOAT and DOUBLE.
• The same TSEncoding.ALP = 15 value in Python, mapped through the C wrapper
  and Cython layer.
• A self-contained ALP page/block wire format with ALP and PLAIN fallback.
• Scalar reference encoder/decoder plus SIMDe AVX2-style kernels.
• SIMD value conversion, SIMD bit unpacking, and word-based packing.
• A two-stage exponent/factor search and direct decoding from wrapped page
  buffers.
• Opt-in integration with the existing EncoderFactory/DecoderFactory.
  The default FLOAT/DOUBLE encoding remains Gorilla.

Java is intentionally untouched. The wire format is language-neutral so a
Java decoder can be added later without changing encoded bytes.

Design

• cpp/docs/alp-codec-design.md describes the
  page/block format, exception handling, and the relationship to TS_2DIFF.
• cpp/docs/alp-vs-gorilla-benchmark.md
  records the ALP/Gorilla/TS_2DIFF comparison.
• cpp/docs/alp-profiling-analysis.md
  records the flame-graph analysis and the optimizations that followed it.

The codec is bit-exact by construction: the encoder decodes each candidate
integer back to the original type and stores an exception when the round-trip
is not bitwise equal. Non-decimal or high-entropy columns fall back to PLAIN;
ALP remains opt-in and should be selected per column.

Benchmark

Codec-level benchmark on Apple Silicon, best of 5, 1M values:

Type	Codec	Ratio	Encode MB/s	Decode MB/s
FLOAT decimal	ALP	1.773	539.3	2646.9
FLOAT decimal	Gorilla	1.457	342.0	1389.3
DOUBLE decimal	ALP	5.882	1055.3	6470.7
DOUBLE decimal	Gorilla	0.970	276.7	2546.8

The benchmark also includes TS_2DIFF as a decimal-data reference. TS_2DIFF has
a much higher ratio on this workload, but the current C++ TS_2DIFF path is not
bit-exact for all FLOAT/DOUBLE values; ALP is the lossless option.

Testing

• Full ./mvnw -P with-cpp ... clean verify passed locally with all bundled
  dependencies enabled: 974/974 tests passed, with only the existing
  disabled/skipped tests not run.
• The CI compatibility build with -Denable.simd=OFF also passed.
• New C++ tests cover scalar round-trip, SIMD/scalar parity, encoder/decoder
  factory integration, special values, truncated payloads, and end-to-end
  TsFile write/read with ALP.
• New Python test verifies TSEncoding.ALP and an ALP FLOAT/DOUBLE
  round-trip through the Cython layer.
• CI passes on Ubuntu, macOS, Windows/MinGW, and Windows/MSVC across the
  supported Python versions; Spotless, cppcheck, RAT, CodeQL, and the
  cross-language fixture validation jobs also pass.

Compatibility

ALP writes a new encoding byte. Older readers that do not know encoding 15
will reject ALP chunks, so the codec is opt-in and the default encodings are
unchanged. Existing files are unaffected.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 86.21795% with 86 lines in your changes missing coverage. Please review.
✅ Project coverage is 64.22%. Comparing base (9b40e40) to head (1762ad2).

Files with missing lines	Patch %	Lines
cpp/src/encoding/alp/alp_decoder.h	43.90%	46 Missing ⚠️
cpp/src/encoding/alp/alp_scalar.h	91.63%	22 Missing ⚠️
cpp/src/encoding/alp/alp_encoder.h	75.67%	9 Missing ⚠️
cpp/src/encoding/alp/alp_simde.h	94.69%	6 Missing ⚠️
cpp/src/encoding/alp/alp_format.h	98.57%	1 Missing ⚠️
cpp/src/encoding/decoder.h	50.00%	1 Missing ⚠️
cpp/src/reader/chunk_reader.cc	0.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##           develop     #966      +/-   ##
===========================================
+ Coverage    63.96%   64.22%   +0.26%     
===========================================
  Files          758      764       +6     
  Lines        53253    53881     +628     
  Branches      8445     8555     +110     
===========================================
+ Hits         34061    34604     +543     
- Misses       17529    17614      +85     
  Partials      1663     1663              

☔ 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.