Run specs in randomized order

[yahonda]

Summary

CONTRIBUTING.md already documents that specs run in randomized order with a reproducible seed, but the configuration to make that true was missing — RSpec ran in :defined order. This PR enables random ordering and fixes the four order-dependent specs that the change surfaces, so the suite is stable under randomization.

The five commits are intentionally split for review:

1. Run specs in randomized order — set config.order = :random and Kernel.srand config.seed in spec/spec_helper.rb; remove the unused RSpec 1.x-era spec/spec.opts (its --loadby mtime --reverse directly contradicts random order); align the CONTRIBUTING.md banner example with RSpec's actual start-of-run output.

2. Reset package state between Numeric variable examples — the "Numeric" shared examples in variable_spec.rb shared numeric_var across three examples; with random order the default-value check could observe a previously-assigned value. Added DBMS_SESSION.RESET_PACKAGE in before(:each) so every example re-initializes to defaults.

3. Avoid mutating shared @Employees in view spec — the :column => nil condition example in view_spec.rb mutated @employees.last in place, leaving the array — built once in before(:all) — corrupted for any subsequent example. Switched to merge so the shared array stays intact.

4. Reset plsql.activerecord_class after ActiveRecord connection examples — schema_spec.rb's ActiveRecord describe bound the global plsql singleton to AR via before(:each) but never reset it. Because oci_connection.ora_date_to_ruby_date consults plsql.default_timezone (a singleton) and that accessor returns ActiveRecord.default_timezone (:utc in modern Rails) when an AR class is bound, every later test reading DATE values back through any OCI connection saw UTC. Added after(:all) { plsql.activerecord_class = nil }.

5. Avoid orphaning the outer session in the Oracle 9.2 nested describe — the nested "using Oracle 9.2" describe in procedure_spec.rb called plsql.connect! again in its before(:all) before the version-skip check. That replaced the outer describe's connection with a fresh session; when skip fired the nested after(:all) was suppressed, but the outer after(:all) still ran on the new session — its drop_session_ruby_temporary_tables filtered by the new session_id and never touched the outer session's ruby_<outer_sid>_<package_id>_* tables, while the outer session itself was leaked. Once that orphan held its temporary tables in use, any later drop_all_ruby_temporary_tables hit ORA-14452. Dropped the redundant plsql.connect!; the version check now uses the outer describe's existing connection.

Test plan

• Verified each failing seed reproduces against current master and is fixed on this branch (variable_spec / view_spec / connection_spec timezone / connection_spec temp tables, seeds 35757 / 45895 / 47562 / 12899).
• Ran the full suite 10 random seeds × 3 trials for each previously failing seed — 0 failures.
• Ran 10 fresh random runs on the integrated branch — 10/10 clean (461 examples, 1 pending, 0 failures).
• CI on this branch (Oracle Database Free 23c, Ruby 4.0).