Fix api/global test flake

[jmacd]

--- FAIL: TestHandlerTestSuite (0.03s)
--- FAIL: TestHandlerTestSuite/TestNoDropsOnDelegate (0.02s)
handler_test.go:129:
Error Trace: handler_test.go:129
Error: Received unexpected error:
no errors sent in 10ms
Test: TestHandlerTestSuite/TestNoDropsOnDelegate
Messages: switched to new Handler
FAIL
FAIL go.opentelemetry.io/otel/api/global 0.082s

[MrAlias]

I think we may need to bump our CI instance size again. I've noticed that error when I run tests locally limiting to a single core.

Also, I've seen this failure under those conditions as well:

--- FAIL: TestNewExporter_collectorConnectionDiesThenReconnects (0.51s)
    otlp_integration_test.go:349: Round #1: Connected collector: spans: got 0 want 10
FAIL
FAIL	go.opentelemetry.io/otel/exporters/otlp	1.556s
ok  	go.opentelemetry.io/otel/exporters/otlp/internal/transform	1.069s
FAIL
Makefile:96: recipe for target 'test' failed
make: *** [test] Error 1

[MrAlias]

also, this

--- FAIL: TestHandlerTestSuite (0.05s)
    --- FAIL: TestHandlerTestSuite/TestNoDropsOnDelegate (0.01s)
        handler_test.go:121: 
            	Error Trace:	handler_test.go:121
            	Error:      	Received unexpected error:
            	            	no errors sent in 10ms
            	Test:       	TestHandlerTestSuite/TestNoDropsOnDelegate
            	Messages:   	starting error stream
FAIL
coverage: 1.5% of statements in ./...
FAIL	go.opentelemetry.io/otel/api/global	0.172s

[MrAlias]

Oof, also, this

ok  	go.opentelemetry.io/otel/sdk/metric/controller/pull	1.156s	coverage: 9.4% of statements in ./...
--- FAIL: TestPushExportError (0.01s)
    --- FAIL: TestPushExportError/errNone (0.00s)
        push_test.go:255: 
            	Error Trace:	push_test.go:255
            	Error:      	Not equal: 
            	            	expected: 1
            	            	actual  : 0
            	Test:       	TestPushExportError/errNone
FAIL
coverage: 10.0% of statements in ./...
FAIL	go.opentelemetry.io/otel/sdk/metric/controller/push	0.104s

[Aneurysm9]

Would this issue be fixed more durably by using GitHub Actions, as proposed in #880?