Call-site recording: empty spans under //line, and one walker shared by L1 and L3 #26

Describe the bug
Two problems in how call sites are recorded:

1. Call nodes under //line directives get empty byte spans (was Call nodes under //line directives get empty byte spans #21). In a file with a //line directive, a call node's positions come from the directive-adjusted line:col, but its byte offsets index the real file. Slicing module.source then returns an empty string, e.g. span.bytes [433,433].
2. L1 and L3 each walk call sites separately (was Share one call-site walker between L1 and L3 #20). L1 records call sites in buildCallSites (internal/syntactic_analysis/symbol_table.go). L3 repeats the same walk in recordedCalls (internal/dataflow/bodynodes.go), and closures are walked twice the same way (buildInnerCallables and closureUnits). L3 matches its calls to L1's call keys by position in that walk. A count check catches a change in the number of calls, but not a change in order with the same count.

To Reproduce

1. Create a Go file containing //line virtual.y:900:1, followed by a function that makes calls.
2. Run cango -i <dir> -a 1 -o /tmp/out.
3. Inspect those call nodes' span.bytes: start equals end.

Expected behavior

• span.bytes covers the call's real text in the file. The line:col can stay directive-adjusted or be real, as long as the docs say which.
• One shared walker yields call sites and closures in recording order for both L1 and L3, so the two cannot drift apart.

Logs
Observed in the testdata/dataflow/linedir fixture added by PR #16.

Additional context

• Both problems predate PR Go L3 intraprocedural dataflow, app-outermost ids and L2 library calls #16. That PR fixed callee resolution under //line; only the span is still wrong.
• Affects generated code: goyacc, ragel, templ and cgo output.
• The span fix lands in the code the shared walker would replace, so do them together.
• Consolidates Share one call-site walker between L1 and L3 #20 and Call nodes under //line directives get empty byte spans #21.