[Xuanwo]

Which issue does this PR close?

Related to #8421's Java/Sled CI failure.

Rationale for this change

The Java behavior extension creates a new Sled operator for each test class but randomizes only the logical key root. Every class still opens the same physical database directory. Two class initializations in the linked job failed with WouldBlock while acquiring the lock on sled/db.

What changes are included in this PR?

• Use the per-initialization UUID for a Sled datadir subdirectory under the configured parent. OPENDAL_DISABLE_RANDOM_ROOT=true preserves the configured root and datadir.
• Add a Sled regression that keeps two fixtures open and writes and reads distinct bytes through both operators. The regression is disabled for other services and when random-root isolation is explicitly disabled.

Validation on nier with the real embedded Sled database:

• The same regression on main df3e3aa6 fails during initialization with the observed database-lock error.
• The fixed full Java/Sled behavior suite reports 56 tests, zero failures or errors, and 36 skips. The new regression executes and passes.
• Native compilation and Java Spotless checks pass.

Are there any user-facing changes?

No. This changes Java behavior-test setup only.

Breaking changes

AI Usage Statement

• Harness: Codex. Model: gpt-6.1-sol. Effort: max. Role: Implemented the fixture change and regression, analyzed CI logs, and prepared validation and this PR.
• Harness: Codex. Model: group/deepseek. Effort: max. Role: Independently reviewed the source and real Sled validation on nier. The original CI lock holder remains unknown; the regression reproduces the collision with two live fixtures.

[PsiACE]

hugs from bub