[ZEPPELIN-6465] Extract magic numbers in EmbeddingSearch.saveIndex() into named constants

[huiseong29]

What is this PR for?

EmbeddingSearch declares its tuning values as named private static final int constants near the top of the class (EMBEDDING_DIM, MAX_TEXT_LENGTH, MAX_INDEX_ENTRIES), but the two truncation limits in saveIndex() were left as bare literals. Each value appeared twice — once in the length check and once in the substring(0, ...) call — so changing only one occurrence would silently let the check bound and the truncation bound drift apart.

This extracts 2000 and 1000 into MAX_PERSISTED_TEXT_LENGTH and MAX_PERSISTED_OUTPUT_LENGTH, following the constant style already used in the class, and replaces all four usages.

The names deliberately distinguish these persistence limits from the existing MAX_TEXT_LENGTH (1500), which bounds in-memory processing before embedding and is a different value for a different purpose. Javadoc on the new text constant makes that distinction explicit.

Values are unchanged, so behavior and the persisted binary format are identical and INDEX_VERSION does not need to move. loadIndex() has no corresponding truncation logic and is untouched.

What type of PR is it?

Refactoring

Todos

• Extract the two truncation limits into named constants
• Replace all four literal usages in saveIndex()

What is the Jira issue?

https://issues.apache.org/jira/browse/ZEPPELIN-6465

How should this be tested?

The EmbeddingSearch tests are gated behind ZEPPELIN_EMBEDDING_TEST and require the embedding model to be installed first:

bash bin/install-search-model.sh
ZEPPELIN_EMBEDDING_TEST=true ./mvnw test -pl zeppelin-server -Dtest='*EmbeddingSearch*'

All 12 tests pass. grep -n "2000\|1000" EmbeddingSearch.java shows the values only at the two constant declarations, with no bare literals left in saveIndex().

Screenshots (if appropriate)

N/A

Questions:

• Does the license files need to update? No
• Is there breaking changes for older versions? No
• Does this needs documentation? No

[huiseong29]

The three failing frontend checks appear unrelated to this change:

• test-selenium-with-spark-module: AuthenticationIT.testSimpleAuthentication
  fails in logoutUser with ElementClickIntercepted — the login modal is still
  fading in and receives the click intended for the user menu.
• run-playwright-e2e-tests (auth): one failure in the paragraph spec on firefox;
  the sibling test in the same file was reported flaky on chromium and passed on retry.
• run-playwright-e2e-tests (anonymous): cancelled as a result of the above.

This PR only extracts two literals into named constants in a Java class, with no
behavior change.

[voidmatcha]

LGTM 👍

The CI failures are known flakes, tracked as issues:

• ZEPPELIN-6711: AuthenticationIT clicks logout before the login modal closes (selenium)
• ZEPPELIN-6710: Monaco suggestions swallow Enter in NB-PARITY-003 (Playwright, already assigned)

ZEPPELIN-6711 is unassigned, so feel free to take it if you're interested.

[ParkGyeongTae]

LGTM

[voidmatcha]

Merged into master (2ee180f).