reduce CGO call overhead for exec and bind paths

[mattn]

Combine multiple CGO crossings into single C calls for hot paths.

• _sqlite3_exec_no_args(): merges prepare+step+finalize into one CGO call for parameterless exec (the most common case)
• _sqlite3_reset_clear(): merges sqlite3_reset + sqlite3_clear_bindings into one CGO call
• Use semaphore channel instead of result-struct channel in context-aware exec/Next
• Use time.AppendFormat with stack buffer to avoid heap allocation when binding time.Time
• Reuse single buffer in bindNamedIndices instead of 3 C.CString allocations
• Remove intermediate bindIndices slice in named parameter binding
• Pass explicit length to sqlite3_prepare_v2 to skip C-side strlen

                                │   before    │    after     │                     │
                                │   sec/op    │   sec/op     │                     │
Suite/BenchmarkExec-16             1.913µ        1.349µ        -29.44% (p=0.000)
Suite/BenchmarkQuery-16            4.977µ        4.488µ         -9.83% (p=0.028)
Suite/BenchmarkParams-16           5.074µ        4.750µ         -6.38% (p=0.001)
Suite/BenchmarkStmt-16             3.077µ        3.005µ              ~ (p=0.169)
Suite/BenchmarkRows-16             130.9µ        130.9µ              ~ (p=0.878)
Suite/BenchmarkStmtRows-16        128.8µ        127.7µ              ~ (p=0.234)
Suite/BenchmarkQueryParallel-16    2.310µ        2.291µ              ~ (p=0.959)
geomean                            8.459µ        7.891µ        -6.72%

                                │    B/op    │    B/op     │
Suite/BenchmarkExec-16            144          72           -50.00% (p=0.000)

                                │ allocs/op  │ allocs/op  │
Suite/BenchmarkExec-16            6            4           -33.33% (p=0.000)