feat: implement LOAD_FILE and AI_SPLIT_DOCUMENT

[ZHEZHEBOY]

Summary

Implement the two DBSchool 2026 Document AI functions requested for seekdb:

• LOAD_FILE(location_name, filename)
• AI_SPLIT_DOCUMENT(content[, params_json])

The change is limited to the parser/resolver/execution/serialization/registration/build wiring needed by those two functions. It does not modify any official test or result file.

Implementation

LOAD_FILE

• Adds scalar expression registration and code generation for LOAD_FILE.
• Resolves a named LOCATION, checks access, builds the object path, enforces document_ai_file_max_size, reads the file, and returns a BLOB.
• Preserves NULL propagation and reports invalid names, missing files, empty files, access errors, and unsupported sizes through existing seekdb error paths.

Core files:

• src/sql/engine/expr/ob_expr_ai/ob_expr_load_file.{h,cpp}
• expression type, factory, evaluator, code-generation, and parameter registration files

AI_SPLIT_DOCUMENT

• Adds MySQL grammar, resolver, statement printing, serialization, and JSON-table execution wiring for the table function.
• Supports text and Markdown input, sentence/word splitting, max, and overlap.
• Returns chunk_id, chunk_offset, chunk_length, and chunk_text.
• Keeps JSON parsing in the SQL layer and passes plain parameter values to oblib, preserving module-layer boundaries and allocator lifetimes.

Core files:

• deps/oblib/src/lib/ai_split_document/*
• src/sql/engine/basic/ob_json_table_op.{h,cpp}
• parser/resolver/printer/item-type/build registration files

Scope review

• Final diff: 25 files, 1,876 insertions, 9 deletions.
• Scope classification: EXPECTED.
• No generated source, test/result changes, unrelated formatting, macOS compatibility code, build artifacts, logs, absolute local paths, credentials, or unrelated features are included.
• The file count is consistent with the required end-to-end SQL plumbing and is smaller than the linked upstream Document AI reference implementation.

Validation

Environment: Ubuntu 22.04, 16 vCPU, targeted build with -j8.

PASS

• bash build.sh init
• bash build.sh debug -DOB_USE_CCACHE=ON -DNEED_PARSER_CACHE=OFF
• make -j8 module_layer_check oblib_lib ob_sql
• make -j8 observer
• built binary revision: 17e03803463db1171f7c0682c16c6ad17972f88e
• observer startup with explicit data-disk resource limits
• test_utility: 1/1
• official mysqltest ai_funcs/load_file: ok
• official mysqltest ai_funcs/ai_split_document: ok
• result-file SHA-256 values unchanged before and after mysqltest
• supplementary runtime checks:
  • empty, short, and Unicode split input
  • invalid max=0 rejection
  • empty and missing-file LOAD_FILE error paths
• git diff --check
• clean tracked server worktree after exact removal of CMake-generated untracked files

FAIL

• Generic test_sql_fast_parser result comparison failed.
• The exact command and output hashes were reproduced on the untouched vldb_2026 baseline (a640cc3), so this is not caused by this PR. No expected result was modified.

NOT RUN

• The complete repository-wide mysqltest/unit-test suite was not run because this PR is intentionally limited to the two Document AI functions and the requested workflow prioritizes targeted validation.
• No test requiring unrelated services or modules was represented as passing.

Remaining validation

Official GitHub Actions are expected to run the repository-wide checks for the PR.

[CLAassistant]

All committers have signed the CLA.

[LINxiansheng]

Document AI & IK Custom Dictionary Score

Document AI Functions Score
===========================
score: 100.00 / 100
load_file: 50 / 50
ai_split_document: 50 / 50

IK Custom Dictionary Score
==========================
score: 0.00 / 100
ik_custom_dict: 0 / 100

FTS Large Benchmark Score

FTS Large Benchmark Score
=========================
score: 4.22 / 100
mean_improvement: 2.11%
full_score_improvement: 50.00%

build_improvement: 1.37%
  build_ik_all_sec: baseline=35.2836, current=35.097, improvement=0.53%
  build_ik_content_sec: baseline=28.3764, current=28.124, improvement=0.89%
  build_beng_en_sec: baseline=14.7578, current=14.362, improvement=2.68%
tokenize_improvement: 3.88%
  tokenize_ik_avg_ms: baseline=0.76478, current=0.7437, improvement=2.76%
  tokenize_beng_avg_ms: baseline=0.42262, current=0.4015, improvement=5.00%
query_improvement: 1.08%
  query_cn_avg_ms: baseline=16.6628, current=16.4331, improvement=1.38%
  query_beng_avg_ms: baseline=24.3042, current=24.1097, improvement=0.80%
  query_mixed_avg_ms: baseline=17.5593, current=17.3522, improvement=1.18%
  query_limit_avg_ms: baseline=16.2334, current=16.0773, improvement=0.96%

FTS Large Benchmark Report

========================================
FTS Large Benchmark Report
========================================
timestamp:              2026-07-14 18:24:55
label:                  vldb-ci-29355666698-1
git_head:               708c302
git_dirty:              0
rows:                   20000
batch:                  500
rounds:                 3000
query_rounds:           200
samples:                3
warmup:                 30
skip_load:              0
----------------------------------------
select1_avg_ms:         0.2072
select1_stdev_ms:       0.0020
raw_load_sec:           1.540
raw_load_rows_per_sec:  12987.0
build_ik_all_sec:       35.097
build_ik_content_sec:   28.124
build_beng_en_sec:      14.362
build_total_sec:        77.596
----------------------------------------
tokenize_ik_avg_ms:     0.7437
tokenize_ik_median_ms:  0.7419
tokenize_ik_stdev_ms:   0.0029
tokenize_beng_avg_ms:   0.4015
tokenize_beng_median_ms:0.4010
tokenize_beng_stdev_ms: 0.0009
----------------------------------------
query_cn_hits:          8001
query_cn_avg_ms:        16.4331
query_cn_stdev_ms:      0.0058
query_beng_hits:        11000
query_beng_avg_ms:      24.1097
query_beng_stdev_ms:    0.0312
query_mixed_hits:       7332
query_mixed_avg_ms:     17.3522
query_mixed_stdev_ms:   0.0486
query_limit_hits:       20
query_limit_avg_ms:     16.0773
query_limit_stdev_ms:   0.0098
========================================

Workflow run