fix: improve behaviour across masking, commands and status (#1194)

Fixes 51 bugs discovered via comparison with `actions/runner`. Every fix has test coverage.

### Secrets

- A short secret registered no shifted-base64 form, so `base64("user:$TOKEN")` printed in the clear
- Encoded forms came only from the whole trimmed value, missing padded and per-line spellings
- Masks split only on `\n`, so `::add-mask::a%0Db` registered neither half
- Adds XML, expression-string and quote-trimming encoders

### Workflow commands

- Split at the last `::` or `]` rather than the first, so `::add-mask::a::b` registered no mask
- A command on the last line without a newline was ignored, and `::ADD-MASK::` did nothing
- `##[...]` did not decode `%3B`/`%5D`, properties lost anything after a second `=`
- `$GITHUB_ENV` and `::set-env::` now refuse `NODE_OPTIONS`

### Status

- `continue-on-error` reported failed, a cancelled job reported success, an `if:` error reported cancelled
- File commands ran after `continue-on-error`, failing the job while the step stayed green
- A bad job output aborted the whole run instead of that job

### Steps and actions

- `${{ matrix.* }}` and `${{ strategy.* }}` were empty inside composite actions
- Composite inputs leaked into nested actions as `INPUT_*`, `with:` matched case-sensitively, `pre` failures were dropped
- Docker actions dropped `runs.env` when the caller passed `with: args:`, and caller `args`/`entrypoint` beat the manifest
- An implicit shell ran with `pipefail`, a `shell:` without `{0}` passed without running
- `container.env` overrode job env and every `$GITHUB_ENV` write, heredocs lost leading blank lines, `$GITHUB_PATH` was not BOM-decoded

Written by Claude Opus 5.

Reviewed-on: https://gitea.com/gitea/runner/pulls/1194
Reviewed-by: bircni <bircni@icloud.com>
Co-authored-by: silverwind <me@silverwind.io>

act/runner/job_executor_test.go

@@ -299,6 +299,7 @@ func TestNewJobExecutor(t *testing.T) {
 		executedSteps []string
 		result        string
 		hasError      bool
+		output        string
 	}{
 		{
 			name:          "zeroSteps",
@@ -319,8 +320,8 @@ func TestNewJobExecutor(t *testing.T) {
 			executedSteps: []string{
 				"startContainer",
 				"step1",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "success",
@@ -336,8 +337,8 @@ func TestNewJobExecutor(t *testing.T) {
 			executedSteps: []string{
 				"startContainer",
 				"step1",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "failure",
@@ -354,8 +355,8 @@ func TestNewJobExecutor(t *testing.T) {
 				"startContainer",
 				"pre1",
 				"step1",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "success",
@@ -372,8 +373,8 @@ func TestNewJobExecutor(t *testing.T) {
 				"startContainer",
 				"step1",
 				"post1",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "success",
@@ -391,8 +392,8 @@ func TestNewJobExecutor(t *testing.T) {
 				"pre1",
 				"step1",
 				"post1",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "success",
@@ -418,13 +419,22 @@ func TestNewJobExecutor(t *testing.T) {
 				"step3",
 				"post3",
 				"post2",
-				"stopContainer",
 				"interpolateOutputs",
+				"stopContainer",
 				"closeContainer",
 			},
 			result:   "success",
 			hasError: false,
 		},
+		{
+			name:          "jobOutputExpressionFailure",
+			steps:         []*model.Step{{ID: "1"}},
+			preSteps:      []bool{false},
+			postSteps:     []bool{false},
+			executedSteps: []string{"startContainer", "step1", "interpolateOutputs", "stopContainer", "closeContainer"},
+			result:        "failure",
+			output:        "${{ 'test' != test }}",
+		},
 	}
 
 	contains := func(needle string, haystack []string) bool {
@@ -450,6 +460,10 @@ func TestNewJobExecutor(t *testing.T) {
 				},
 				Config: &Config{},
 			}
+			if tt.output != "" {
+				rc.Run.Job().Outputs = map[string]string{"bad": tt.output}
+				rc.outputTemplate = map[string]string{"bad": tt.output}
+			}
 			rc.ExprEval = rc.NewExpressionEvaluator(ctx)
 			executorOrder := make([]string, 0)
 
@@ -497,6 +511,9 @@ func TestNewJobExecutor(t *testing.T) {
 
 				jim.On("interpolateOutputs").Return(func(ctx context.Context) error {
 					executorOrder = append(executorOrder, "interpolateOutputs")
+					if tt.output != "" {
+						return rc.interpolateOutputs()(ctx)
+					}
 					return nil
 				})
 
@@ -518,6 +535,7 @@ func TestNewJobExecutor(t *testing.T) {
 			executor := newJobExecutor(jim, sfm, rc)
 			err := executor(ctx)
 			assert.NoError(t, err) //nolint:testifylint // pre-existing issue from nektos/act
+			assert.Empty(t, rc.Run.Job().Outputs["bad"])
 			assert.Equal(t, tt.executedSteps, executorOrder)
 
 			jim.AssertExpectations(t)